Predicting positive youth development among Chinese adolescents: A machine learning approach using multiwave longitudinal data

Zelin Liu; Zékai Lu; Yaqiong Wang; Danhua Lin

doi:10.1111/aphw.70211

. 2026 Sep 4;18(5):e70211. doi: 10.1111/aphw.70211

Predicting positive youth development among Chinese adolescents: A machine learning approach using multiwave longitudinal data

Zelin Liu ¹, Zékai Lu ^1,², Yaqiong Wang ³, Danhua Lin ^1,^✉

PMCID: PMC13545513 PMID: 42698279

Abstract

Despite growing recognition that positive youth development (PYD) depends on the dynamic interaction of individual and ecological resources, existing studies rely on linear models that cannot capture high‐dimensional, nonlinear predictor configurations. This study applied machine learning to four‐wave longitudinal data from 5019 Chinese adolescents (ages 9–19) to identify the key predictors of PYD at T4 (controlling for prior PYD at T3), measured by the Chinese 4Cs model (Character, Competence, Confidence, Connection). We compared 12 algorithms; CatBoost achieved the best prediction ( $R^{2}$ = .816). SHAP analysis identified school psychological climate, depression, and parental loneliness as the top three predictors. Heterogeneity analyses revealed an age gradient: School climate dominated for primary and middle school students, whereas parental loneliness dominated for high school students. Student type analyses uncovered three distinct developmental pathways: an aspirational pathway characterized by social mobility belief for migrant children, a relational pathway characterized by parental loneliness for left‐behind and urban children, and a clinical pathway characterized by depression for rural ordinary children. These findings provide empirical support for differentiated, context‐sensitive intervention strategies targeting PYD across diverse Chinese adolescent populations.

Keywords: Chinese adolescents, heterogeneity analysis, longitudinal prediction, machine learning, positive youth development, SHAP

INTRODUCTION

Globally, the healthy development of adolescents has become one of the central issues in developmental psychology and public health. The World Health Organization has emphasized that adolescence is a critical period for establishing lifelong health foundations, and that investing in adolescent health and development concerns individual lifelong well‐being and constitutes a core driver of sustainable societal progress (World Health Organization, 2024). Against this macro background, the field of developmental psychology has undergone a major paradigm shift, moving from the traditional “deficit” and “psychopathology” perspective that focused on identifying psychological and behavioral problems toward a more strength‐based approach: the positive youth development (PYD) perspective (Lerner, Lerner, et al., 2015). This perspective critiques the conventional assumption that “the absence of problems equals health,” arguing that “problem‐free is not fully prepared” and that holistic health should encompass the flourishing (thriving) of positive psychological attributes. Accordingly, adolescents should not be viewed merely as “problems to be managed” but rather as “resources to be developed” (Roth & Brooks‐Gunn, 2003), with an emphasis on achieving comprehensive development through constructive interactions between individuals and their multiple ecological environments.

It is worth noting that the meaning of PYD is profoundly shaped by cultural context. In Western theoretical frameworks, PYD is commonly operationalized through the “Five Cs” model, comprising Competence, Confidence, Connection, Character, and Caring (Lerner et al., 2005). In Eastern cultures, particularly those influenced by Confucianism, positive adolescent development emphasizes three interrelated aspects: the realization of individual competence, social connection, and moral character (Character and Connection; Shek et al., 2013). On this basis, indigenous research has constructed a culturally adapted PYD framework for the Chinese context, the “4Cs” model, encompassing Character, Competence, Confidence, and Connection (Chai et al., 2022). Although the PYD concept and its value as a “firewall” against subsequent mental health problems have gained broad recognition (The Lancet Psychiatry, 2025), the empirical evidence regarding which key factors can longitudinally and dynamically predict PYD trajectories remains insufficient, calling for deeper investigation.

From a theoretical standpoint, the mechanisms underlying positive adolescent development are grounded in relational developmental systems (RDS), the overarching framework for contemporary PYD models (Lerner et al., 2019). RDS posits that human development is fundamentally driven by continuous, bidirectional, and dynamic interactions between an inherently active individual and their multiple, integrated ecological environments (i.e. individual ⇔ context relations; Lerner, Lerner, et al., 2015). Building on this metatheoretical foundation, developmental contextualism further highlights the “optimal match” between individual strengths (e.g. individual agency) and changing ecological resources (Lerner et al., 2006). To operationalize this dynamic process, developmental assets theory specifies the concrete resources involved in this process, proposing that positive development stems from the joint contribution of external ecological resources (e.g. supportive family and school environments) and internal personal resources (e.g. skills and values) (Benson et al., 2011). These resources exhibit “horizontal stacking” effects across contexts at a single time point and “vertical stacking” dynamics that accumulate over time (Benson, 2007). Taken together, to accurately characterize and predict PYD trajectories, prediction models must move beyond a single, static perspective and incorporate comprehensive variable sets spanning multiple time points and multiple ecological systems (individual, family, school, and macro‐social), thereby capturing both the cumulative effects and dynamic interactions of developmental resources.

Despite the theoretical clarity that the cumulative and dynamic interactions of multisystemic variables are central to adolescent development, empirical PYD research faces significant methodological challenges. The vast majority of studies rely heavily on traditional linear regression or structural equation modeling (SEM; Årdal et al., 2018; Milot Travers & Mahalik, 2021). These conventional statistical methods are typically built on strict linearity assumptions and encounter limitations involving multicollinearity, model convergence, and computational complexity when fitting large samples, making it difficult to simultaneously process and analyze high‐dimensional predictor variables (De Souza‐Lima et al., 2025; Pujadas et al., 2025). More critically, within the complex ecological system of adolescent development, the effects of individual factors and external environmental resources on positive development are often not simply additive in a linear fashion; they may exhibit diminishing returns or threshold effects (Gao et al., 2025). Constrained by prespecified model structures, traditional methods are prone to overlooking these intricate nonlinear patterns and interaction effects (Gao et al., 2025), thereby limiting our precise characterization and understanding of the complex formation mechanisms of PYD.

In the face of these limitations, the introduction of machine learning (ML) provides a paradigmatic methodological breakthrough for PYD research. Compared with traditional statistical models, ML demonstrates unparalleled advantages in processing high‐dimensional complex data, capturing nonlinear interactions, and identifying deep data patterns (Gao et al., 2025; Jiabin et al., 2025). In recent years, ML has achieved notable successes in psychological research: Researchers have used algorithms such as random forests, XGBoost (Chen & Guestrin, 2016), and CatBoost (Prokhorenkova et al., 2017) to successfully predict adolescent subjective well‐being, social–emotional skills, and mental health outcomes (Jiabin et al., 2025; Meng et al., 2025) and have found that ML models explain significantly more variance than traditional regression (De Souza‐Lima et al., 2025). More importantly, modern interpretable ML tools such as SHAP (Lundberg et al., 2020; Ponce‐Bobadilla et al., 2024; Salih et al., 2025) have successfully opened the algorithmic “black box,” substantially improving predictive accuracy and precisely revealing the contribution weights, effect directions, and complex threshold effects of specific predictors (Long et al., 2025; Pelt et al., 2024). Although ML has achieved initial successes in predicting mental health risk, almost no studies have used ML to systematically investigate PYD predictors, and research that combines multiwave, multidimensional longitudinal data to analyze the dynamic mechanisms of PYD development is extremely scarce.

To address this methodological gap and construct a comprehensive feature space for ML modeling, it is essential to first identify the broad spectrum of potential predictors. Extensive systematic reviews highlight that PYD is profoundly shaped by a complex interplay of individual capabilities and contextual affordances (Ardoin et al., 2022; Deane & Dutton, 2020). Grounded in developmental assets theory and RDS theory, empirical research consistently identifies internal psychological resources (e.g. resilience) and external ecological resources (e.g. teacher support and positive school climates) as the key predictors of youth thriving (Martin‐Barrado & Gomez‐Baya, 2024; Yuan & Liu, 2025). However, predicting developmental trajectories requires moving beyond a narrow focus on protective assets. As highlighted by the intersection of adolescent health and resilience literatures, demographic characteristics (Gomez‐Baya et al., 2022; Wang et al., 2025) and physical lifestyle factors (Almeida et al., 2025; Curran & Wexler, 2017) serve as fundamental biological and behavioral foundations that moderate psychosocial development (Deane & Dutton, 2020). Furthermore, from a dynamic systems perspective (Huang et al., 2026), positive attributes and psychopathological risks (e.g. depression and anxiety) function as interactive, competing systems; the accumulation of risk factors can erode developmental assets over time, making it impossible to fully understand thriving without accounting for co‐occurring vulnerabilities (Shek et al., 2021). To systematically capture this multisystemic complexity, the present study organizes potential predictors of PYD into five theoretically grounded domains: (a) demographics and personality, representing foundational individual traits; (b) internal psychological resources, reflecting core personal strengths; (c) external family and school resources, capturing primary ecological supports; (d) physical and lifestyle factors, representing behavioral health foundations; and (e) risk factors, accounting for the competing vulnerabilities that may hinder positive trajectories.

Crucially, the longitudinal impacts of these multidomain factors are rarely uniform across all adolescents. The predictive weights of developmental resources often fluctuate dynamically across different age groups (Chai & Lin, 2021; Grasmeijer et al., 2024). Moreover, the mechanisms fostering positive development exhibit significant heterogeneity across diverse socio‐cultural subgroups. In the context of rapid urbanization in China, structurally marginalized populations—such as left‐behind children and migrant children—face unique developmental barriers and family dynamics that fundamentally alter their access to and utilization of ecological resources (Wang et al., 2025; Wen et al., 2015). Therefore, to precisely characterize the formation of PYD, predictive models must not only assess the longitudinal accumulation of multidomain variables but also explicitly examine the heterogeneity among different age groups and student types. This differentiated perspective is essential for moving beyond a “one‐size‐fits‐all” assumption, thereby informing targeted, context‐sensitive intervention strategies for diverse adolescent populations.

To address these gaps, this study applied ML to multiwave longitudinal data from Chinese adolescents, systematically identifying key predictors of PYD. We employed a three‐layer analytical framework: (1) comparing 12 algorithms to select the optimal model, (2) using SHAP to interpret predictor contributions, and (3) conducting heterogeneity analyses across age groups and student types (e.g. left‐behind and migrant children). This study represents the first systematic application of ML and interpretability methods to the longitudinal prediction of PYD among Chinese adolescents, aiming to move beyond the linearity constraints of traditional methods and reveal the key factors and nonlinear patterns predicting PYD development. The findings will provide solid empirical evidence and data‐driven decision support for building precision‐targeted, individualized intervention systems for positive adolescent development.

The Present Study

Drawing on developmental assets theory and RDS theory, we assembled 100 predictor variables spanning five ecological domains (demographics and personality, internal psychological resources, external family and school resources, physical and lifestyle factors, and risk factors) measured across three time points (T1–T3) to predict PYD at T4 among 5019 Chinese adolescents aged 9–19 (Figure 1).

Conceptual framework of predictor variables organized by ecological domain. *Note*: Using a tree metaphor, the outcome (PYD at T4) represents the “Fruits” of development, nourished by demographic “Soil,” internal psychological “Roots,” external ecological “Sunlight,” physical and lifestyle “Leaves,” and risk factor “Lightning.” Variables marked with an asterisk (*) were measured at multiple time points (T1–T3). (**) were measured at multiple time points (T2–T3).

Our analytical framework has three layers. First, we compare 12 ML algorithms to identify the optimal predictive model. Second, we apply SHAP to the best model to quantify the importance, direction, and nonlinear effects of each predictor. Third, we conduct heterogeneity analyses to examine whether the predictive architecture of PYD differs across age groups and student types defined by migration status. Three research questions guide the analysis:

Which ML algorithm provides the most accurate prediction of longitudinal PYD at T4, and how do ensemble tree‐based methods compare to linear and instance‐based approaches?
Which ecological and psychological factors are the most important incremental predictors of PYD at T4 (controlling for prior [T3] PYD levels), and what are the directions and nonlinear patterns of their effects?
Does the predictive architecture of PYD vary across age groups and student types defined by migration status?

DATA AND METHODS

Participants and Data

Data were drawn from a four‐wave longitudinal study tracking Chinese adolescents across four time points (T1–T4). The baseline survey was conducted in fall 2017 (T1), with follow‐ups in fall 2018 (T2), spring 2019 (T3), and fall 2019 (T4). Thus, the T1–T2 interval was approximately 12 months, whereas the T2–T3 and T3–T4 intervals were approximately 6 months each. Participants were recruited from six schools (two elementary, two middle, and two high schools) in each of three provinces (Jiangxi, Liaoning, and Henan), selected to reflect variation in economic development, geographic location, and educational quality. The baseline sample comprised 5019 participants aged 9–19 years, with slightly more female (54.4%) than male participants (45.6%). Table 1 presents the full sample demographics.

TABLE 1.

Sample demographics (N = 5019).

Variable

Measures

Outcome Variable

The outcome was PYD at T4, measured using the Chinese 4Cs model (Chai et al., 2020, 2022), which comprises four subdimensions: Character, Competence, Confidence, and Connection. A composite PYD score was computed as the mean of the four subdimension scores. Supplementary multigroup confirmatory factor analyses supported configural, metric, and scalar measurement invariance of the T4 PYD measure across gender, age groups, and student types (see Table S8). PYD at T3 was included as a covariate (autoregressive control) to ensure that the model predicted change in PYD from T3 to T4 rather than T4 levels alone.

Predictor variables

We assembled 100 predictor variables spanning multiple ecological domains and three time points (T1–T3), organized into the following categories. Key instruments are described below; full measurement details including psychometric properties are reported in Table S1.

Demographics and personality traits

Gender, age, grade, student type, subjective socioeconomic status (MacArthur Scale; Adler et al., 2000), Big Five personality traits (TIPI; Gosling et al., 2003), and temperament (emotionality and activity).

Internal psychological resources

Social mobility belief measured adolescents' conviction that personal effort can lead to upward social mobility, a construct particularly relevant to China's stratified educational and economic system. Future control assessed perceived mastery over future outcomes. Resilience was measured using the Connor–Davidson Resilience Scale (CD‐RISC; Connor & Davidson, 2003), which assesses adaptive coping capacity under adversity. Intentional self‐regulation captured goal‐directed behavioral regulation. Emotion regulation was assessed via two subscales: cognitive reappraisal (reframing the meaning of emotional situations) and expressive suppression (inhibiting outward emotional expression). Prosocial behavior was measured using the relevant subscale of the Strengths and Difficulties Questionnaire (SDQ). Academic performance and student leadership (class officer status) were assessed via single items.

Family and parental factors

Parental educational involvement, which measures the degree to which parents actively engage in their child's educational activities, was assessed using an adapted scale (Cheung & Pomerantz, 2015). Parent–child communication was measured using the Acculturative Family Distancing (AFD) scale, which captures the quality and openness of communication between adolescents and their parents. Parental conflict was decomposed into three dimensions: frequency, intensity, and resolution, using the Children's Perception of Interparental Conflict Scale (CPIC). Parental loneliness, a key variable in this study, was measured using the parental subscale of the Marcoen Loneliness Scale (Marcoen et al., 1987), which assesses perceived emotional distance between adolescents and their parents; higher scores indicate greater feelings of emotional disconnection from parents. Parental marriage status and parental education level (father and mother separately) were assessed as single items.

School and community context

School psychological climate, key school status, public vs. private school, and regional economic indicators (area‐level GDP, Gini coefficient, average wage, per capita consumption, disposable income, green coverage rate).

Risk factors and health behaviors

Depression (CES‐D; Radloff, 1977), general and peer loneliness, externalizing problems, adolescent life events, stress‐related physical symptoms, screen time, physical exercise, sleep duration and disturbance (PSQI), and nutritional intake.

Missing Data

Missing values in predictor variables were imputed using MissForest (Stekhoven & Bühlmann, 2012), a nonparametric, random‐forest‐based imputation method that handles mixed‐type data (continuous and categorical) without distributional assumptions. Imputation was applied to the full predictor matrix before the train‐test split. Although performing imputation within the cross‐validation loop would be methodologically preferable, the low missing‐data rates in this study (0%–12% across variables) limit the scope for information leakage. The outcome variable (PYD at T4) was not imputed; cases with missing outcome values were excluded from analysis. Additional details regarding the rationale for using MissForest, convergence, and imputation diagnostics are provided in Supporting Information S2.

ANALYTIC STRATEGY

The analysis proceeded in four stages. First, we compared 12 ML algorithms encompassing three families: regularized linear models (Lasso, Ridge, ElasticNet), ensemble tree‐based methods (Random Forest, ExtraTrees, Gradient Boosting, AdaBoost, XGBoost, LightGBM, CatBoost), and instance‐based/kernel methods (K‐Nearest Neighbors, Support Vector Regression). All models were trained on an 80/20 train‐test split with 5‐fold cross‐validation on the training set. Performance was evaluated using $R^{2}$ , root mean square error (RMSE), and mean absolute error (MAE) on the held‐out test set.

Second, gradient boosting algorithms—CatBoost, LightGBM, and XGBoost—underwent Bayesian hyperparameter optimization using Optuna with 100 iterations per model. The optimization targeted learning rate, tree depth, regularization strength, subsampling ratios, and algorithm‐specific parameters (e.g. bagging temperature for CatBoost and number of leaves for LightGBM). The model with the highest test $R^{2}$ was designated as the final model. The analysis followed a two‐stage procedure: In Stage 1, all 12 algorithms were trained with default hyperparameters and evaluated on the test set, and the three gradient boosting algorithms were selected for tuning based on their shared architectural suitability; in Stage 2, Bayesian optimization with 5‐fold cross‐validation was conducted exclusively on the training set, with the test set held out entirely.

Third, we applied SHAP TreeExplainer (Lundberg & Lee, 2017) to the final model to compute feature importance. SHAP values decompose each prediction into additive contributions from each feature (Lundberg et al., 2020; Lundberg & Lee, 2017), providing both the magnitude and direction of each predictor's effect. PYD at T3 was included in the model as a control but excluded from SHAP displays, so that all reported feature importances reflect incremental predictive contributions beyond prior PYD levels.

Fourth, we conducted two sets of heterogeneity analyses. (a) Age group analysis: Separate models were fit for primary school (9–12), middle school (13–15), and high school (16–19) students. (b) Student type analysis: Separate models were fit for migrant, left‐behind, rural ordinary, and urban ordinary children. For each subgroup, CatBoost, XGBoost, and LightGBM were trained with globally tuned hyperparameters, and the best‐performing model was selected. The Bayesian optimization and cross‐validation procedures are described in Supporting Information S3, and the hyperparameter search spaces are reported in Table S2.

RESULTS

Model Comparison and Selection

Table 2 presents the predictive performance of all 12 algorithms. Ensemble tree‐based models consistently outperformed regularized linear models and instance‐based methods. Among the default (untuned) models, CatBoost achieved the highest test $R^{2}$ of .815 (RMSE = .402), followed by ExtraTrees ( $R^{2}$ = .808), Gradient Boosting ( $R^{2}$ = .806), and LightGBM ( $R^{2}$ = .806). The three regularized linear models (Ridge, ElasticNet, Lasso) clustered around $R^{2}$ = .795, whereas SVR (.777), AdaBoost (.744), and KNN (.704) performed worst. The narrow performance gap between the top six ensemble methods (range: .803–.815) indicates that the underlying signal is stable across algorithmic choice within the tree‐based family.

TABLE 2.

Model performance.

Model	CV $R^{2}$	Test $R^{2}$	RMSE	MAE
Panel A: Default models (sorted by test R ²)
CatBoost	.806	.815	.402	—
ExtraTrees	.799	.808	.410	—
Gradient Boosting	.800	.806	.411	—
LightGBM	.801	.806	.411	—
XGBoost	.798	.805	.413	—
Random Forest	.795	.803	.415	—
Ridge	.791	.796	.422	—
ElasticNet	.793	.795	.423	—
Lasso	.793	.794	.424	—
SVR	.779	.777	.442	—
AdaBoost	.744	.744	.472	—
KNN	.715	.704	.508	—
Panel B: After Bayesian hyperparameter tuning
CatBoost*	.807	.816	.401	.288
LightGBM	.808	.813	.404	.292
XGBoost	.807	.811	.406	.295

Open in a new tab

Note: CV = 5‐fold cross‐validation on the training set (N = 4015). * = final selected model.

After Bayesian hyperparameter tuning of the three gradient boosting algorithms, CatBoost improved to $R^{2}$ = .816 (RMSE = .401, MAE = .288), LightGBM to .813 (RMSE = .404), and XGBoost to .811 (RMSE = .406). CatBoost was selected as the final model for all subsequent full‐sample analyses. The strong predictive performance of ensemble tree‐based models is consistent with recent comparative studies in behavioral science (Jiabin et al., 2025; Pelt et al., 2024). The close agreement between cross‐validation and test‐set performance across all models (differences were approximately .01 or lessin $R^{2}$ ) suggests minimal overfitting, likely due to the large sample size relative to the feature space (N = 5019 with 100 features). The overall explanatory power ( $R^{2}$ = .816) indicates that the feature predictor set, combined with prior PYD levels (T3), accounts for approximately 82% of the variance in T4 PYD, representing strong predictive performance for a developmental outcome measured approximately six months after T3.

Global Feature Importance

Figure 2 presents the SHAP beeswarm and bar plots for the tuned CatBoost model. The top three predictors, after controlling for prior PYD (T3), were school psychological climate at T3 (mean|SHAP| = .057), depression at T3 (|SHAP| = .056), and parental loneliness at T3 (|SHAP| = .043). Parental loneliness at T2 ranked fourth (|SHAP| = .042), and social mobility belief at T2 ranked fifth (.038). Resilience at T1 (.028), school psychological climate at T2 (.027), and depression at T2 (.026) comprised the next tier. The complete ranking of the top 25 features by mean absolute SHAP value is provided in Table S3.

Global SHAP analysis. *Note*: PYD (T3) controlled for but excluded from display; all importances reflect incremental predictive contributions. Top 15 features shown. Panel (a) Bee swarm plot. Each point represents one observation; color indicates feature value (red = high, blue = low). Panel (b) Mean absolute SHAP values.

Several patterns are notable in the beeswarm plot. School psychological climate showed a clear positive gradient: Higher values of school climate at T3 were associated with positive SHAP contributions (i.e. increases in PYD), whereas lower values pushed PYD downward. Depression showed the inverse pattern, with higher depression scores consistently linked to negative SHAP contributions. Parental loneliness showed a similar risk‐factor pattern: Greater parental loneliness at both T2 and T3 predicted PYD decline. Social mobility belief displayed a positive relationship, with higher beliefs about upward social mobility contributing positively to PYD. Resilience at T1 functioned as a long‐range protective factor, with its influence persisting across the full T1‐to‐T4 interval.

The top 15 predictors spanned three ecological domains: school context (school psychological climate at T1, T2, T3), family relationships (parental loneliness at T2 and T3, parental educational involvement at T3, parent–child communication at T2), and individual psychological resources (depression at T2 and T3, social mobility belief at T2, resilience at T1). This multidomain profile is consistent with ecological systems theory, developmental assets theory, and RDS theory, which predict that PYD emerges from the joint contribution of internal and external resources across ecological systems.

Two temporal patterns deserve attention. First, several predictors appeared at multiple time points (e.g. school climate at T1, T2, and T3; depression at T2 and T3; and parental loneliness at T2 and T3), with later waves carrying greater predictive weight. This temporal gradient suggests cumulative or intensifying effects over time, consistent with the “vertical stacking” principle of developmental assets theory (Benson, 2007). However, because the assessment intervals were unequal, the greater predictive weight of later‐wave measures may partly reflect their closer temporal proximity to T4. Therefore, this pattern should not be interpreted solely as evidence of developmental accumulation. Second, resilience at T1 was the only T1‐measured psychological variable in the top 10, indicating that early resilience retained predictive importance across the approximately 2‐year T1–T4 interval, even after controlling for more proximal predictors. This pattern supports the conceptualization of resilience as a foundational developmental asset rather than a state‐dependent resource.

Age Group Heterogeneity

Age‐group‐specific models (Figure 3) uncovered a developmental gradient in PYD predictors. All three age‐group models performed well: middle school ( $R^{2}$ = .845, CatBoost), high school ( $R^{2}$ = .841, XGBoost), and primary school ( $R^{2}$ = .742, CatBoost). Detailed subgroup sample sizes, held‐out test‐set sizes, best‐performing models, and predictive performance are reported in Table S4, whereas the complete top‐25 SHAP rankings for the three age groups are presented in Table S5.

SHAP bee swarm plots by age group. *Note*: PYD (T3) controlled for but excluded from display. Top 10 features shown. Panel (a) Primary school (9–12); Panel (b) Middle school (13–15); Panel (c) High school (16–19).

For primary school students (ages 9–12), school psychological climate at T3 was the top predictor (|SHAP| = .051), followed by parental loneliness at T2 (.048) and depression at T3 (.039). For middle school students (ages 13–15), school psychological climate at T3 remained dominant (.056), with depression at T3 (.051) and parental loneliness at T3 (.045) close behind. A qualitative shift appeared for high school students (ages 16–19): Parental loneliness at T3 rose sharply to the top position (.125), exceeding the second‐ranked predictor, depression at T3 (.078), by 60%. School psychological climate dropped to rank 4 (.045).

This pattern indicates a developmental transition. During early and mid‐adolescence, school constitutes the primary developmental context, with school climate as the top predictor of PYD. As adolescents enter late adolescence and face university entrance examinations and the transition to adulthood, parent–child emotional bonds become the dominant factor. This age gradient is consistent with attachment theory, which posits that adolescents facing high‐stress transitions reactivate their reliance on parents as a “safe haven” and “secure base” (Delgado et al., 2022).

Figure 4 visualizes this developmental transition by tracking how the rank of each top‐15 feature shifts across age groups. The crossover between school psychological climate (declining from rank 1 in primary school to rank 4 in high school) and parental loneliness (rising from rank 10 in primary school to rank 1 in high school) is clearly visible, as is the steady rise of depression across age groups.

Feature importance rank shifts across age groups (top 15 features). *Note*: Lower rank indicates greater importance.

Feature importance rank shifts across age groups (top 15 features). The crossover between school psychological climate and parental loneliness illustrates the developmental transition from school‐dominated to family‐dominated predictive architecture.

Student Type Heterogeneity

Student‐type‐specific models (Figure 5) revealed three distinct developmental pathways. Predictive accuracy varied across groups: rural ordinary ( $R^{2}$ = .838, LightGBM), left‐behind ( $R^{2}$ = .817, XGBoost), urban ordinary ( $R^{2}$ = .789, LightGBM), and migrant ( $R^{2}$ = .747, XGBoost). Detailed subgroup sample sizes, held‐out test‐set sizes, best‐performing models, and predictive performance are reported in Table S6, whereas the complete top‐25 SHAP rankings for the four student types are presented in Table S7.

SHAP bee swarm plots by student type. *Note*: PYD (T3) controlled for but excluded from display. Top 10 features shown. Panel (a) Migrant; Panel (b) Left‐behind; Panel (c) Rural ordinary; Panel (d) Urban ordinary.

Migrant children (n = 1126) showed a unique predictor profile. Social mobility belief at T2 was the top predictor (|SHAP| = .066), followed by depression at T3 (.059), school psychological climate at T3 (.057), parent–child communication at T3 (.047), parental loneliness at T3 (.044), future control at T3 (.040), and parental loneliness at T2 (.040). The dominance of social mobility belief, the conviction that effort can overcome structural disadvantage, reveals an aspirational pathway specific to migrant children. Unlike other subgroups where emotional or relational factors dominate, migrant children's PYD is most strongly predicted by their beliefs about upward mobility, a finding that reflects the motivational context of rural‐to‐urban migration. Depression at rank 2 and school psychological climate at rank 3 capture the emotional burden and institutional context of migration‐related adjustment, whereas parent–child communication (rank 4) and parental loneliness (ranks 5 and 7) reflect the importance of family relational quality during transitions.

Left‐behind children (n = 875) were dominated by parental loneliness at T3 (.098), followed by social mobility belief at T2 (.068), school psychological climate at T3 (.052), and social mobility belief at T3 (.043). The prominence of parental loneliness, which was more than 40% larger than the next predictor, represents a relational pathway (Wen et al., 2015) driven by the emotional cost of parental absence due to labor migration. That social mobility belief ranked second indicates that left‐behind children's PYD also depends on maintaining aspirational beliefs about their future despite current family separation (Xiang et al., 2024).

Rural ordinary children (n = 772) showed a clinical pathway, with depression at T3 as the dominant predictor (.086), followed by parental loneliness at T3 (.067) and depression at T2 (.047). The outsized role of depression in this group likely reflects the compounding of limited mental health resources and greater psychosocial stressors in rural communities.

Urban ordinary children (n = 2246) most closely resembled the overall sample profile. Parental loneliness at T3 ranked first (.091), followed by depression at T3 (.060), school psychological climate at T3 (.047), and parental loneliness at T2 (.032). Parent–child communication at T2 (.029) and parental educational involvement at T3 (.028) also appeared in the top ranks. This relational pathway mirrors the left‐behind group, though the mechanism differs: For urban children, parental loneliness reflects the quality of parent–child emotional engagement in intact families rather than the structural absence of parents. The broader spread of SHAP values across family‐related predictors (four of the top 10 were family variables) suggests that urban children's PYD is shaped by a constellation of family relational factors rather than any single dominant influence.

The variation in optimal model selection across student types is itself informative. XGBoost performed best for migrant and left‐behind children, whereas LightGBM performed best for rural and urban ordinary children. This algorithmic heterogeneity suggests that the functional form of predictor–outcome relationships differs across populations, reinforcing the value of separate subgroup analyses rather than a single pooled model with interaction terms.

DISCUSSION

This study is the first to use ML methods to systematically identify the key predictors of longitudinal PYD development among Chinese adolescents. By comparing 12 ML algorithms, we found that ensemble tree‐based models demonstrated strong performance in predicting PYD change. The results revealed that school positive psychological climate (T3) is the most important external resource for promoting positive adolescent development; depression (T3) and parental loneliness (T3/T2) are the two primary risk factors inhibiting individual flourishing; and social mobility belief (T2) and early resilience (T1) are key internal assets predicting positive development. Beyond these overall patterns, positive adolescent development is not a homogeneous, “one‐size‐fits‐all” process but rather exhibits significant heterogeneity. Specifically, the model revealed a clear age gradient: The leading predictors of PYD gradually shift from the school environment to family relationships as individuals mature. The study also found that PYD predictors are population‐specific; for instance, the PYD predictors of left‐behind children (parental loneliness) and migrant children (social mobility belief) depend on distinctly different ecological and psychological resources.

School Psychological Climate as the Core External Resource

As the most important external developmental resource, school positive psychological climate (T3) was the most prominent incremental predictor of PYD change. From the perspective of developmental assets theory, the school, as the core ecological system for adolescents' daily activities, is the primary setting for the provision of external developmental resources and directly determines whether adolescents can effectively access social support, build positive interpersonal relationships, and accumulate adaptive developmental capital (Benson et al., 2011). Our finding is highly consistent with prior research. For example, Gao et al. (2025), using decision tree analysis, found that school belonging was the single most important predictor distinguishing levels of adolescent well‐being. Similarly, ML research on high‐risk adolescent populations has also shown that school‐related factors are among the most important predictors of subsequent mental health outcomes (Pujadas et al., 2025). This suggests that optimizing school psychological climate can most directly and effectively promote PYD development.

Depression and Parental Loneliness as Key Risk Factors

This study revealed that depressive symptoms (T3) and parental loneliness (T2/T3) constitute the two core risk factors impeding PYD development. First, depressive symptoms (T3) ranked second, and their negative predictive effect indicates that depression is not merely a “problem” of concern to psychopathology but a formidable “barrier” blocking the path to adolescent flourishing. Our finding aligns with the growing body of ML‐based well‐being research, which consistently identifies mental health variables as key predictors of developmental outcomes (Long et al., 2025; Pujadas et al., 2025). Recent empirical work has also confirmed that depression, as a core symptom within the pathological system, continuously “scars” the positive development system of adolescents (Huang et al., 2026). This suggests that in efforts to promote PYD, preventing and providing early intervention for adolescent depressive symptoms must be treated as foundational work.

Second, parental loneliness consistently ranked among the top PYD predictors across multiple time points (ranked fourth and third, respectively), which shows the sustained and powerful cross‐temporal influence of the family emotional environment on adolescent development. From the perspective of RDS theory (Lerner, Johnson, & Buckingham, 2015), stable, positive emotional bonds between adolescents and their significant others, especially parents, represent an important relational resource associated with positive development. When adolescents chronically perceive emotional disconnection from their parents, this psychological “resource deprivation” significantly undermines the accumulation of internal developmental capital. This finding is consistent with existing literature on the importance of parent–child emotional bonds for adolescent adjustment (Delgado et al., 2022; Wen & Lin, 2012).

Social Mobility Belief and Resilience as Internal Developmental Assets

Internal psychological resources and individual agency play an equally irreplaceable role in predicting PYD trajectories. On one hand, social mobility belief (T2) constitutes an exceptionally valuable internal developmental asset for adolescents. From the RDS perspective, adolescents are not passive “sponges” absorbing environmental influences; rather, they shape their own development through active interaction with their environments (Lerner, Johnson, & Buckingham, 2015). Adolescents with strong social mobility beliefs tend to possess greater intrinsic motivation to actively acquire, optimize, and create environmental resources. This aligns with existing empirical research on the positive association between social mobility beliefs and mental health (Lin et al., 2022; Xiang et al., 2024).

On the other hand, this study further confirmed the long‐term predictive value of early (T1) resilience. As a core dynamic system capacity enabling positive adaptation under adversity, early resilience provides adolescents with a foundational “floor of developmental protection” (Masten et al., 2021). Empirical research has also demonstrated that resilience is an important predictor of well‐being (Gao et al., 2025) and helps individuals buffer the negative impact of stressful events (Klainin‐Yobas et al., 2021; Masten et al., 2021). These findings collectively indicate that cultivating early social mobility beliefs and resilience in adolescents is key to activating their endogenous developmental motivation and promoting positive development.

Age‐Graded Developmental Transition

The age heterogeneity analysis revealed a significant “age gradient” transition in the key contextual resources for positive adolescent development: As individuals mature, the core predictors shift from the school context toward family emotional support. This finding strongly corroborates the central tenet of developmental contextualism, namely, that the key contexts and resource needs for individual development dynamically shift with age and changing developmental tasks (Lerner et al., 2006). During early and middle adolescence, the school, as the primary arena for adolescent socialization and activities, directly provides individuals with a sense of belonging and normative guidance (Korpershoek et al., 2020), constituting the core “normative resource” for positive development (Oberle, 2018). When adolescents enter later high school years, however, facing the intense pressures and high uncertainties of college entrance examinations and the transition to adulthood, individuals tend to rely more heavily on parents' emotional availability. As attachment theory suggests, when coping with external stress and unknown challenges, adolescents still require the family as a psychological “safe haven” and a “secure base” for exploration (Delgado et al., 2022; Gültekin Ahçı & Akdeniz, 2025); once insecure attachment patterns form, they may trigger problematic behaviors (Scott et al., 2011). This further suggests that PYD intervention practice must adopt a developmental, dynamic perspective. In the high school stage in particular, the focus of educational and clinical interventions should tilt toward repairing and strengthening the deep emotional bond between parents and adolescents.

Student Type Heterogeneity and Three Developmental Pathways

This study revealed that positive adolescent development exhibits high specificity across groups with different migration statuses and urban–rural backgrounds, profoundly reflecting the differentiated impact of China's distinctive urban–rural dual structure on individual development. Specifically, for migrant children who relocated with their parents to cities, social mobility belief (T2) ranked first among PYD predictors. Constrained by structural disadvantages such as the household registration (hukou) system, migrant children often face challenges including discrimination and peer victimization in urban areas (Nie et al., 2022; Xiang et al., 2024); yet, the intrinsic motivation and belief that “effort can change one's social class” emerged as their strongest positive predictor of subsequent PYD (Lin et al., 2022).

For left‐behind children, parental loneliness (T3) surged to the top of the predictor ranking. This result directly reflects the physical absence and emotional deprivation caused by prolonged parental migration for work, indicating that mending the rupture in the parent–child bond and alleviating family‐level emotional deprivation is strongly associated with addressing the positive development challenges of this group (Wang et al., 2025; Wen & Lin, 2012).

Among rural ordinary children, depressive symptoms (T3) were the foremost bottleneck impeding PYD development. This phenomenon reflects the reality that in rural areas with relatively scarce medical and quality educational psychological resources, untreated emotional distress and psychological difficulties continuously and intensely deplete individuals' psychological capital (Wang et al., 2025).

In sum, future intervention practices targeting Chinese adolescent PYD must abandon a “one‐size‐fits‐all” approach and instead provide precision‐targeted, individualized support programs for special adolescent groups in different ecological environments.

Theoretical and Practical Implications

The findings of this study carry important value for both theoretical deepening and practical guidance. At the theoretical level, this study is the first to use interpretable ML technology to empirically verify and deepen the core hypotheses of RDS theory and developmental assets theory using large‐scale longitudinal data. The research provides strong data support for the cross‐cultural applicability of these Western developmental psychology theories within an Eastern Confucian cultural context but also precisely reveals the complex nonlinear patterns and high specificity of key PYD predictors across different age stages and special populations.

At the practical level, this study provides clear empirical support for building precision‐targeted, individualized intervention systems for positive adolescent development. First, resource allocation and intervention targets should be precision‐oriented, focusing on the primary core objectives: comprehensively cultivating a positive, supportive school climate; establishing routine adolescent depression screening and early intervention mechanisms; and substantively enhancing the quality of parent–child emotional bonds. Second, intervention strategies must abandon homogeneity and instead deliver precision interventions based on adolescents' specific social‐ecological backgrounds: For migrant children, educators and communities should strengthen role‐model inspiration and career planning education to bolster social mobility beliefs; for left‐behind children and urban ordinary children, there is an urgent need to develop family‐based intervention programs that strengthen parent–child bonds and alleviate loneliness. The setting of intervention priorities should also follow the “age gradient” pattern: Primary and middle school stages should focus on school climate building, whereas the high school stage should shift the intervention focus toward repairing and deepening family parent–child relationships.

Limitations and Future Directions

Despite the breakthrough findings, this study has several limitations. First, although ML models offer advantages in capturing high‐dimensional nonlinear patterns, they fundamentally identify complex associative relationships that cannot be directly equated with rigorous causal inference. Future research should integrate causal ML algorithms or quasi‐experimental designs to further confirm the causal pathways between core predictors and PYD trajectories. Second, the study primarily relied on adolescent self‐report, which may have introduced common method variance, recall bias, and social desirability bias. Future research could attempt to incorporate multimodal, multisource data such as parent and teacher assessments and physiological and behavioral logs to improve the objectivity and comprehensiveness of measurement. Third, although this study leveraged a large‐scale, multiwave sample, the data remain limited to the single cultural and social context of China. Future research should validate the prediction model across countries and cultural groups to examine its generalizability and culture‐specific patterns. Fourth, the student‐type‐specific analyses did not assess interindividual differences in intraindividual PYD trajectories. Future research could use longitudinal growth or person‐centered trajectory models to examine developmental heterogeneity. Fifth, subgroup‐specific SHAP rankings should be interpreted cautiously because they may be affected by unequal sample sizes, model instability, and correlated predictors. Finally, MissForest imputation before the train‐test split may have introduced minor information leakage. Although missingness was relatively low (0%–12%), future studies should perform imputation within cross‐validation.

CONCLUSION

In sum, this study is the first to systematically apply ML methods with large‐scale longitudinal data to construct and interpret a dynamic prediction model for PYD among Chinese adolescents. The findings reveal that school positive psychological climate, as the most important external resource, together with depressive symptoms and parental loneliness as core risk factors, and social mobility belief and early resilience as key internal capital, collectively form a multidimensional, dynamically interactive developmental prediction system. Of particular importance, this study uncovered significant heterogeneity in PYD development: The leading predictors shifted from the school context toward family emotional support with increasing age, displaying an “age gradient” transition. At the same time, the key predictors of positive development for different groups, including migrant children, left‐behind children, and rural children, exhibit high specificity.

In the face of the challenges of the modern social ecology, we wish to call attention to the fact that adolescents are by no means “problems to be managed” but rather “resources to be developed” that urgently need investment and cultivation. Only by precisely identifying and providing matching multilevel ecological support can we truly safeguard the positive development and lifelong well‐being of every adolescent.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

ETHICS STATEMENT

This study was approved by the Ethics Committee of the Faculty of Psychology, Beijing Normal University (Approval No. BNU001912210084). All procedures performed in this study were in accordance with the ethical standards of the institutional research committee and with the Declaration of Helsinki and its later amendments. Participants were clearly informed about the research purpose, procedures, potential risks, and their right to withdraw at any time without consequences. All participants provided consent for the use of their anonymized data in academic publications. All data collected were treated with strict confidentiality, and any personally identifiable information was anonymized to ensure privacy.

Supporting information

Table S1: Complete List of Variables, Measurement Instruments, and References.

Table S2: Hyperparameter Search Spaces.

Table S3: SHAP Feature Rankings.

Table S4: Model Performance ‐ Age Groups.

Table S5: Top 25 SHAP Features by Age Group.

Table S6: Model Performance — Student Types.

Table S7: Top 25 SHAP Features by Student Type.

Table S8: Measurement Invariance of T4 Positive Youth Development Across Gender, Age Groups, and Student Types.

APHW-18-0-s001.docx^{(67.1KB, docx)}

ACKNOWLEDGMENTS

We extend our sincere gratitude to the adolescents who participated in this multiwave longitudinal study. Their sustained engagement and willingness to contribute their time and experiences were indispensable to this research. We are also deeply thankful to the scientific research personnel whose expertise and dedication in data collection and management made this investigation possible.

Liu, Z. , Lu, Z. , Wang, Y. , & Lin, D. (2026). Predicting positive youth development among Chinese adolescents: A machine learning approach using multiwave longitudinal data. Applied Psychology: Health and Well‐Being, 18(5), e70211. 10.1111/aphw.70211

Funding information This study was supported by the National Natural Science Foundation of China (32471116).

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

REFERENCES

Adler, N. E. , Epel, E. S. , Castellazzo, G. , & Ickovics, J. R. (2000). Relationship of subjective and objective social status with psychological and physiological functioning: Preliminary data in healthy, white women. Health Psychology, 19(6), 586–592. 10.1037/0278-6133.19.6.586 [DOI] [PubMed] [Google Scholar]
Almeida, L. , Dias, T. , Corte‐Real, N. , Menezes, I. , & Fonseca, A. (2025). Positive youth development through sport and physical education: A systematic review of empirical research conducted with grade 5 to 12 children and youth. Physical Education and Sport Pedagogy, 30(3), 282–308. 10.1080/17408989.2023.2230208 [DOI] [Google Scholar]
Årdal, E. , Holsen, I. , Diseth, Å. , & Larsen, T. (2018). The five Cs of positive youth development in a school context; gender and mediator effects. School Psychology International, 39(1), 3–21. 10.1177/0143034317734416 [DOI] [Google Scholar]
Ardoin, N. M. , Bowers, A. W. , Kannan, A. , & O'Connor, K. (2022). Positive youth development outcomes and environmental education: A review of research. International Journal of Adolescence and Youth, 27(1), 475–492. 10.1080/02673843.2022.2147442 [DOI] [Google Scholar]
Benson, P. L. (2007). Developmental assets: An overview of theory, research, and practice. In Approaches to positive youth development (pp. 33–58). Sage. 10.4135/9781446213803.n2 [DOI] [Google Scholar]
Benson, P. L. , Scales, P. C. , & Syvertsen, A. K. (2011). The contribution of the developmental assets framework to positive youth development theory and practice. In Advances in child development and behavior (Vol. 41, pp. 197–230). Elsevier. 10.1016/B978-0-12-386492-5.00008-7 [DOI] [PubMed] [Google Scholar]
Chai, X. , Li, X. , Cao, J. , & Lin, D. (2020). A short form of the Chinese Positive Youth Development Scale: Development and validation in a large sample. Studies of Psychology and Behavior, 18(5), 631–637. [Google Scholar]
Chai, X. , & Lin, D. (2021). School transition during adolescence: Turning crisis into opportunity. Advances in Psychological Science, 29(5), 864–874. 10.3724/SP.J.1042.2021.00864 [DOI] [Google Scholar]
Chai, X. , Wang, J. , Li, X. , Liu, W. , Zhao, G. , & Lin, D. (2022). Development and validation of the Chinese Positive Youth Development Scale. Applied Developmental Science, 26(1), 127–140. 10.1080/10888691.2020.1712206 [DOI] [Google Scholar]
Chen, T. , & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785–794. 10.1145/2939672.2939785 [DOI]
Cheung, C. S.‐S. , & Pomerantz, E. M. (2015). Value development underlies the benefits of parents' involvement in children's learning: A longitudinal investigation in the United States and China. Journal of Educational Psychology, 107(1), 309–320. 10.1037/a0037458 [DOI] [PMC free article] [PubMed] [Google Scholar]
Connor, K. M. , & Davidson, J. R. T. (2003). Development of a new resilience scale: The Connor‐Davidson Resilience Scale (CD‐RISC). Depression and Anxiety, 18(2), 76–82. 10.1002/da.10113 [DOI] [PubMed] [Google Scholar]
Curran, T. , & Wexler, L. (2017). School‐based positive youth development: A systematic review of the literature. Journal of School Health, 87(1), 71–80. 10.1111/josh.12467 [DOI] [PubMed] [Google Scholar]
De Souza‐Lima, J. , Ferrari, G. , Yáñez‐Sepúlveda, R. , Giakoni‐Ramírez, F. , Muñoz‐Strale, C. , Alarcon‐Aguilar, J. , Parra‐Saldias, M. , Duclos‐Bastias, D. , Godoy‐Cumillaf, A. , Merellano‐Navarro, E. , Bruneau‐Chávez, J. , & Valdivia‐Moral, P. (2025). Prediction of children's subjective well‐being from physical activity and sports participation using machine learning techniques: Evidence from a multinational study. Children, 12(8), 1083. 10.3390/children12081083 [DOI] [PMC free article] [PubMed] [Google Scholar]
Deane, K. , & Dutton, H. (2020). The factors that influence positive youth development and wellbeing. Key Concepts, Contemporary International Debates and a Review of the Issues for Aotearoa New Zealand Research, Policy and Practice. Ara Taiohi. https://ir.canterbury.ac.nz/bitstreams/95943194-33aa-48e1-aeca-09c80e09dadc/download
Delgado, E. , Serna, C. , Martínez, I. , & Cruise, E. (2022). Parental attachment and peer relationships in adolescence: A systematic review. International Journal of Environmental Research and Public Health, 19(3), 1064. 10.3390/ijerph19031064 [DOI] [PMC free article] [PubMed] [Google Scholar]
Gao, X. , Wang, C. , & Liu, Y. (2025). From empty to flourishing: Latent profile and decision tree analysis of well‐being in Chinese adolescents. Child Indicators Research, 18(5), 1973–1998. 10.1007/s12187-025-10282-2 [DOI] [Google Scholar]
Gomez‐Baya, D. , Santos, T. , & Gaspar De Matos, M. (2022). Developmental assets and positive youth development: An examination of gender differences in Spain. Applied Developmental Science, 26(3), 516–531. 10.1080/10888691.2021.1906676 [DOI] [Google Scholar]
Gosling, S. D. , Rentfrow, P. J. , & Swann, W. B. (2003). A very brief measure of the big‐five personality domains. Journal of Research in Personality, 37(6), 504–528. 10.1016/S0092-6566(03)00046-1 [DOI] [Google Scholar]
Grasmeijer, A. J. , Gomez‐Baya, D. , Camacho, C. , & Mendoza‐Berjano, R. (2024). The association between family dynamics and Positive Youth Development in secondary education students. Scandinavian Journal of Psychology, 65(5), 858–869. 10.1111/sjop.13028 [DOI] [PubMed] [Google Scholar]
Gültekin Ahçı, Z. , & Akdeniz, S. (2025). Adult parental attachment and adolescent psychological problems: Mediating role of adult emotional processes in the family. The American Journal of Family Therapy, 53(2), 218–240. 10.1080/01926187.2024.2431334 [DOI] [Google Scholar]
Huang, S. , Lei, H. , Yang, Z. , Lai, X. , Wang, X. , Zhang, C. , & Wang, Y. (2026). Competition between the positive youth development and the psychopathological system among adolescents: Less vulnerability or more scarring? Journal of Child Psychology and Psychiatry, and Allied Disciplines, 67(1), 55–66. 10.1111/jcpp.70015 [DOI] [PMC free article] [PubMed] [Google Scholar]
Jiabin, L. , Shanshan, Z. , Xiaomei, L. , Kaixin, Y. , & Ying, Y. (2025). The benefits of physical literacy for human flourishing: A machine learning–based exploration of adolescents. Applied Psychology: HEALTH and Well‐Being, 17(1), e12635. 10.1111/aphw.12635 [DOI] [PubMed] [Google Scholar]
Klainin‐Yobas, P. , Vongsirimas, N. , Ramirez, D. Q. , Sarmiento, J. , & Fernandez, Z. (2021). Evaluating the relationships among stress, resilience and psychological well‐being among young adults: A structural equation modelling approach. BMC Nursing, 20(1), 119. 10.1186/s12912-021-00645-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
Korpershoek, H. , Canrinus, E. T. , Fokkens‐Bruinsma, M. , & De Boer, H. (2020). The relationships between school belonging and students' motivational, social‐emotional, behavioural, and academic outcomes in secondary education: A meta‐analytic review. Research Papers in Education, 35(6), 641–680. 10.1080/02671522.2019.1615116 [DOI] [Google Scholar]
Lerner, R. M. , Johnson, S. K. , & Buckingham, M. H. (2015). Relational developmental systems‐based theories and the study of children and families: Lerner and Spanier (1978) revisited. Journal of Family Theory & Review, 7(2), 83–104. 10.1111/jftr.12067 [DOI] [Google Scholar]
Lerner, R. M. , Lerner, J. V. , Almerigi, J. B. , Theokas, C. , Phelps, E. , Gestsdottir, S. , Naudeau, S. , Jelicic, H. , Alberts, A. , Ma, L. , Smith, L. M. , Bobek, D. L. , Richman‐Raphael, D. , Simpson, I. , Christiansen, E. D. , & Von Eye, A. (2005). Positive youth development, participation in community youth development programs, and community contributions of fifth‐grade adolescents: Findings from the first wave of the 4‐H study of positive youth development. Journal of Early Adolescence, 25(1), 17–71. 10.1177/0272431604272461 [DOI] [Google Scholar]
Lerner, R. M. , Lerner, J. V. , Almerigi, J. , & Theokas, C. (2006). Dynamics of individual ←→ context relations in human development: A developmental systems perspective. In Comprehensive handbook of personality and psychopathology, vol. 1: personality and everyday functioning (pp. 23–43). John Wiley & Sons, Inc. [Google Scholar]
Lerner, R. M. , Lerner, J. V. , Bowers, E. P. , & Geldhof, G. J. (2015). Positive youth development and relational‐developmental‐systems. In Handbook of child psychology and developmental science (pp. 1–45). Wiley. 10.1002/9781118963418.childpsy116 [DOI] [Google Scholar]
Lerner, R. M. , Tirrell, J. M. , Dowling, E. M. , Geldhof, G. J. , Gestsdóttir, S. , Lerner, J. V. , King, P. E. , Williams, K. , Iraheta, G. , & Sim, A. T. R. (2019). The end of the beginning: Evidence and absences studying positive youth development in a global context. Adolescent Research Review, 4(1), 1–14. 10.1007/s40894-018-0093-4 [DOI] [Google Scholar]
Lin, L. , Hua, L. , & Li, J. (2022). Seeking pleasure or growth? The mediating role of happiness motives in the longitudinal relationship between social mobility beliefs and well‐being in college students. Personality and Individual Differences, 184, 111170. 10.1016/j.paid.2021.111170 [DOI] [Google Scholar]
Long, Q. , Wang, Y. , Axelin, A. , Zheng, F. , Cao, Z. , Li, X. , & Guo, J. (2025). Enhancing happiness and life satisfaction in university students: Analysis with a machine learning approach. BMC Psychology, 14(1), 45. 10.1186/s40359-025-03809-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
Lundberg, S. , & Lee, S.‐I. (2017). A unified approach to interpreting model predictions. arXiv (Cornell University). 10.48550/ARXIV.1705.07874 [DOI]
Lundberg, S. M. , Erion, G. , Chen, H. , DeGrave, A. , Prutkin, J. M. , Nair, B. , Katz, R. , Himmelfarb, J. , Bansal, N. , & Lee, S.‐I. (2020). From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2(1), 56–67. 10.1038/s42256-019-0138-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
Marcoen, A. , Goossens, L. , & Caes, P. (1987). Lonelines in pre‐through late adolescence: Exploring the contributions of a multidimensional approach. Journal of Youth and Adolescence, 16(6), 561–577. 10.1007/BF02138821 [DOI] [PubMed] [Google Scholar]
Martin‐Barrado, A. D. , & Gomez‐Baya, D. (2024). A scoping review of the research evidence of the developmental assets model in Europe. Frontiers in Psychology, 15, 1407338. 10.3389/fpsyg.2024.1407338 [DOI] [PMC free article] [PubMed] [Google Scholar]
Masten, A. S. , Lucke, C. M. , Nelson, K. M. , & Stallworthy, I. C. (2021). Resilience in development and psychopathology: Multisystem perspectives. Annual Review of Clinical Psychology, 17(1), 521–549. 10.1146/annurev-clinpsy-081219-120307 [DOI] [PubMed] [Google Scholar]
Meng, H. , He, S. , Guo, J. , Wang, H. , & Tang, X. (2025). Applying machine learning to understand the role of social–emotional skills on subjective well‐being and physical health. Applied Psychology: HEALTH and Well‐Being, 17(1), e12624. 10.1111/aphw.12624 [DOI] [PubMed] [Google Scholar]
Milot Travers, A. S. , & Mahalik, J. R. (2021). Positive youth development as a protective factor for adolescents at risk for depression and alcohol use. Applied Developmental Science, 25(4), 322–331. 10.1080/10888691.2019.1634569 [DOI] [Google Scholar]
Nie, W. , Gao, L. , & Cui, K. (2022). Bullying victimization and mental health among migrant children in urban China: A moderated mediation model of school belonging and resilience. International Journal of Environmental Research and Public Health, 19(12), 7135. 10.3390/ijerph19127135 [DOI] [PMC free article] [PubMed] [Google Scholar]
Oberle, E. (2018). Early adolescents' emotional well‐being in the classroom: The role of personal and contextual assets. Journal of School Health, 88(2), 101–111. 10.1111/josh.12585 [DOI] [PubMed] [Google Scholar]
Pelt, D. H. M. , Habets, P. C. , Vinkers, C. H. , Ligthart, L. , Van Beijsterveldt, C. E. M. , Pool, R. , & Bartels, M. (2024). Building machine learning prediction models for well‐being using predictors from the exposome and genome in a population cohort. Nature Mental Health, 2(10), 1217–1230. 10.1038/s44220-024-00294-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
Ponce‐Bobadilla, A. V. , Schmitt, V. , Maier, C. S. , Mensing, S. , & Stodtmann, S. (2024). Practical guide to SHAP analysis: Explaining supervised machine learning model predictions in drug development. Clinical and Translational Science, 17(11), e70056. 10.1111/cts.70056 [DOI] [PMC free article] [PubMed] [Google Scholar]
Prokhorenkova, L. , Gusev, G. , Vorobev, A. , Dorogush, A. V. , & Gulin, A. (2017). CatBoost: Unbiased boosting with categorical features. arXiv (Cornell University), 31, 6639–6649. 10.48550/ARXIV.1706.09516 [DOI] [Google Scholar]
Pujadas, E. R. , Díaz‐Caneja, C. M. , Stevanovic, D. , Quintero, M. F. , Martín‐Isla, C. , Hernández‐González, J. , Atehortúa, A. , Lazrak, N. , Pries, L. , Delespaul, P. , Camacho, M. , Gülöksüz, S. , Rutten, B. P. F. , & Lekadir, K. (2025). Longitudinal prediction of mental health outcomes in vulnerable youth using machine learning. Cognitive Computation, 17(5), 152. 10.1007/s12559-025-10509-y [DOI] [Google Scholar]
Radloff, L. S. (1977). The CES‐D scale: A self‐report depression scale for research in the general population. Applied Psychological Measurement, 1(3), 385–401. 10.1177/014662167700100306 [DOI] [Google Scholar]
Roth, J. L. , & Brooks‐Gunn, J. (2003). What exactly is a youth development program? Answers from research and practice. Applied Developmental Science, 7(2), 94–111. 10.1207/S1532480XADS0702_6 [DOI] [Google Scholar]
Salih, A. M. , Raisi‐Estabragh, Z. , Galazzo, I. B. , Radeva, P. , Petersen, S. E. , Lekadir, K. , & Menegaz, G. (2025). A perspective on explainable artificial intelligence methods: SHAP and LIME. Advanced Intelligent Systems, 7(1), 2400304. 10.1002/aisy.202400304 [DOI] [Google Scholar]
Scott, S. , Briskman, J. , Woolgar, M. , Humayun, S. , & O'Connor, T. G. (2011). Attachment in adolescence: Overlap with parenting and unique prediction of behavioural adjustment. Journal of Child Psychology and Psychiatry, 52(10), 1052–1062. 10.1111/j.1469-7610.2011.02453.x [DOI] [PubMed] [Google Scholar]
Shek, D. T. L. , Yu, L. , & Fu, X. (2013). Confucian virtues and Chinese adolescent development: A conceptual review. Ijamh, 25(4), 335–344. 10.1515/ijamh-2013-0031 [DOI] [PubMed] [Google Scholar]
Shek, D. T. , Zhao, L. , Dou, D. , Zhu, X. , & Xiao, C. (2021). The impact of positive youth development attributes on posttraumatic stress disorder symptoms among Chinese adolescents under COVID‐19. Journal of Adolescent Health, 68(4), 676–682. 10.1016/j.jadohealth.2021.01.011 [DOI] [PubMed] [Google Scholar]
Stekhoven, D. J. , & Bühlmann, P. (2012). MissForest—Non‐parametric missing value imputation for mixed‐type data. Bioinformatics, 28(1), 112–118. 10.1093/bioinformatics/btr597 [DOI] [PubMed] [Google Scholar]
The Lancet Psychiatry . (2025). New perspectives on prevention. The Lancet Psychiatry, 12(10), 723. 10.1016/S2215-0366(25)00281-0 [DOI] [PubMed] [Google Scholar]
Wang, Y. , Su, S. , Xiao, J. , & Lin, D. (2025). Positive character and psychological well‐being: A nuanced examination among adolescents from diverse rural–urban and migration contexts. Journal of Adolescence, 97(1), 180–195. 10.1002/jad.12409 [DOI] [PubMed] [Google Scholar]
Wen, M. , & Lin, D. (2012). Child development in rural China: Children left behind by their migrant parents and children of nonmigrant families. Child Development, 83(1), 120–136. 10.1111/j.1467-8624.2011.01698.x [DOI] [PubMed] [Google Scholar]
Wen, M. , Su, S. , Li, X. , & Lin, D. (2015). Positive youth development in rural China: The role of parental migration. Social Science and Medicine, 132, 261–269. 10.1016/j.socscimed.2014.07.051 [DOI] [PMC free article] [PubMed] [Google Scholar]
World Health Organization . (2024). The adolescent health indicators recommended by the global action for measurement of adolescent health. World Health Organization. https://books.google.com/books?id=A6cOEQAAQBAJ [Google Scholar]
Xiang, S. , Zhao, S. , Xiao, J. , Wang, Y. , Li, Y. , Liu, Z. , & Lin, D. (2024). Longitudinal trajectories of social mobility beliefs among Chinese adolescents: The protective roles of parental academic involvement and adolescent future orientation. Journal of Youth and Adolescence, 53(8), 1903–1917. 10.1007/s10964-024-01984-x [DOI] [PubMed] [Google Scholar]
Yuan, R. , & Liu, M. (2025). Promoting positive growth through building developmental assets for maltreated adolescents: A multiple mediation analysis. Child Abuse Review, 34(3), e70044. 10.1002/car.70044 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Table S1: Complete List of Variables, Measurement Instruments, and References.

Table S2: Hyperparameter Search Spaces.

Table S3: SHAP Feature Rankings.

Table S4: Model Performance ‐ Age Groups.

Table S5: Top 25 SHAP Features by Age Group.

Table S6: Model Performance — Student Types.

Table S7: Top 25 SHAP Features by Student Type.

Table S8: Measurement Invariance of T4 Positive Youth Development Across Gender, Age Groups, and Student Types.

APHW-18-0-s001.docx^{(67.1KB, docx)}

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

[aphw70211-bib-0001] Adler, N. E. , Epel, E. S. , Castellazzo, G. , & Ickovics, J. R. (2000). Relationship of subjective and objective social status with psychological and physiological functioning: Preliminary data in healthy, white women. Health Psychology, 19(6), 586–592. 10.1037/0278-6133.19.6.586 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0002] Almeida, L. , Dias, T. , Corte‐Real, N. , Menezes, I. , & Fonseca, A. (2025). Positive youth development through sport and physical education: A systematic review of empirical research conducted with grade 5 to 12 children and youth. Physical Education and Sport Pedagogy, 30(3), 282–308. 10.1080/17408989.2023.2230208 [DOI] [Google Scholar]

[aphw70211-bib-0003] Årdal, E. , Holsen, I. , Diseth, Å. , & Larsen, T. (2018). The five Cs of positive youth development in a school context; gender and mediator effects. School Psychology International, 39(1), 3–21. 10.1177/0143034317734416 [DOI] [Google Scholar]

[aphw70211-bib-0004] Ardoin, N. M. , Bowers, A. W. , Kannan, A. , & O'Connor, K. (2022). Positive youth development outcomes and environmental education: A review of research. International Journal of Adolescence and Youth, 27(1), 475–492. 10.1080/02673843.2022.2147442 [DOI] [Google Scholar]

[aphw70211-bib-0005] Benson, P. L. (2007). Developmental assets: An overview of theory, research, and practice. In Approaches to positive youth development (pp. 33–58). Sage. 10.4135/9781446213803.n2 [DOI] [Google Scholar]

[aphw70211-bib-0006] Benson, P. L. , Scales, P. C. , & Syvertsen, A. K. (2011). The contribution of the developmental assets framework to positive youth development theory and practice. In Advances in child development and behavior (Vol. 41, pp. 197–230). Elsevier. 10.1016/B978-0-12-386492-5.00008-7 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0007] Chai, X. , Li, X. , Cao, J. , & Lin, D. (2020). A short form of the Chinese Positive Youth Development Scale: Development and validation in a large sample. Studies of Psychology and Behavior, 18(5), 631–637. [Google Scholar]

[aphw70211-bib-0008] Chai, X. , & Lin, D. (2021). School transition during adolescence: Turning crisis into opportunity. Advances in Psychological Science, 29(5), 864–874. 10.3724/SP.J.1042.2021.00864 [DOI] [Google Scholar]

[aphw70211-bib-0009] Chai, X. , Wang, J. , Li, X. , Liu, W. , Zhao, G. , & Lin, D. (2022). Development and validation of the Chinese Positive Youth Development Scale. Applied Developmental Science, 26(1), 127–140. 10.1080/10888691.2020.1712206 [DOI] [Google Scholar]

[aphw70211-bib-0010] Chen, T. , & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785–794. 10.1145/2939672.2939785 [DOI]

[aphw70211-bib-0011] Cheung, C. S.‐S. , & Pomerantz, E. M. (2015). Value development underlies the benefits of parents' involvement in children's learning: A longitudinal investigation in the United States and China. Journal of Educational Psychology, 107(1), 309–320. 10.1037/a0037458 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0012] Connor, K. M. , & Davidson, J. R. T. (2003). Development of a new resilience scale: The Connor‐Davidson Resilience Scale (CD‐RISC). Depression and Anxiety, 18(2), 76–82. 10.1002/da.10113 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0013] Curran, T. , & Wexler, L. (2017). School‐based positive youth development: A systematic review of the literature. Journal of School Health, 87(1), 71–80. 10.1111/josh.12467 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0014] De Souza‐Lima, J. , Ferrari, G. , Yáñez‐Sepúlveda, R. , Giakoni‐Ramírez, F. , Muñoz‐Strale, C. , Alarcon‐Aguilar, J. , Parra‐Saldias, M. , Duclos‐Bastias, D. , Godoy‐Cumillaf, A. , Merellano‐Navarro, E. , Bruneau‐Chávez, J. , & Valdivia‐Moral, P. (2025). Prediction of children's subjective well‐being from physical activity and sports participation using machine learning techniques: Evidence from a multinational study. Children, 12(8), 1083. 10.3390/children12081083 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0015] Deane, K. , & Dutton, H. (2020). The factors that influence positive youth development and wellbeing. Key Concepts, Contemporary International Debates and a Review of the Issues for Aotearoa New Zealand Research, Policy and Practice. Ara Taiohi. https://ir.canterbury.ac.nz/bitstreams/95943194-33aa-48e1-aeca-09c80e09dadc/download

[aphw70211-bib-0016] Delgado, E. , Serna, C. , Martínez, I. , & Cruise, E. (2022). Parental attachment and peer relationships in adolescence: A systematic review. International Journal of Environmental Research and Public Health, 19(3), 1064. 10.3390/ijerph19031064 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0017] Gao, X. , Wang, C. , & Liu, Y. (2025). From empty to flourishing: Latent profile and decision tree analysis of well‐being in Chinese adolescents. Child Indicators Research, 18(5), 1973–1998. 10.1007/s12187-025-10282-2 [DOI] [Google Scholar]

[aphw70211-bib-0018] Gomez‐Baya, D. , Santos, T. , & Gaspar De Matos, M. (2022). Developmental assets and positive youth development: An examination of gender differences in Spain. Applied Developmental Science, 26(3), 516–531. 10.1080/10888691.2021.1906676 [DOI] [Google Scholar]

[aphw70211-bib-0019] Gosling, S. D. , Rentfrow, P. J. , & Swann, W. B. (2003). A very brief measure of the big‐five personality domains. Journal of Research in Personality, 37(6), 504–528. 10.1016/S0092-6566(03)00046-1 [DOI] [Google Scholar]

[aphw70211-bib-0020] Grasmeijer, A. J. , Gomez‐Baya, D. , Camacho, C. , & Mendoza‐Berjano, R. (2024). The association between family dynamics and Positive Youth Development in secondary education students. Scandinavian Journal of Psychology, 65(5), 858–869. 10.1111/sjop.13028 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0021] Gültekin Ahçı, Z. , & Akdeniz, S. (2025). Adult parental attachment and adolescent psychological problems: Mediating role of adult emotional processes in the family. The American Journal of Family Therapy, 53(2), 218–240. 10.1080/01926187.2024.2431334 [DOI] [Google Scholar]

[aphw70211-bib-0022] Huang, S. , Lei, H. , Yang, Z. , Lai, X. , Wang, X. , Zhang, C. , & Wang, Y. (2026). Competition between the positive youth development and the psychopathological system among adolescents: Less vulnerability or more scarring? Journal of Child Psychology and Psychiatry, and Allied Disciplines, 67(1), 55–66. 10.1111/jcpp.70015 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0023] Jiabin, L. , Shanshan, Z. , Xiaomei, L. , Kaixin, Y. , & Ying, Y. (2025). The benefits of physical literacy for human flourishing: A machine learning–based exploration of adolescents. Applied Psychology: HEALTH and Well‐Being, 17(1), e12635. 10.1111/aphw.12635 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0024] Klainin‐Yobas, P. , Vongsirimas, N. , Ramirez, D. Q. , Sarmiento, J. , & Fernandez, Z. (2021). Evaluating the relationships among stress, resilience and psychological well‐being among young adults: A structural equation modelling approach. BMC Nursing, 20(1), 119. 10.1186/s12912-021-00645-9 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0025] Korpershoek, H. , Canrinus, E. T. , Fokkens‐Bruinsma, M. , & De Boer, H. (2020). The relationships between school belonging and students' motivational, social‐emotional, behavioural, and academic outcomes in secondary education: A meta‐analytic review. Research Papers in Education, 35(6), 641–680. 10.1080/02671522.2019.1615116 [DOI] [Google Scholar]

[aphw70211-bib-0026] Lerner, R. M. , Johnson, S. K. , & Buckingham, M. H. (2015). Relational developmental systems‐based theories and the study of children and families: Lerner and Spanier (1978) revisited. Journal of Family Theory & Review, 7(2), 83–104. 10.1111/jftr.12067 [DOI] [Google Scholar]

[aphw70211-bib-0028] Lerner, R. M. , Lerner, J. V. , Almerigi, J. B. , Theokas, C. , Phelps, E. , Gestsdottir, S. , Naudeau, S. , Jelicic, H. , Alberts, A. , Ma, L. , Smith, L. M. , Bobek, D. L. , Richman‐Raphael, D. , Simpson, I. , Christiansen, E. D. , & Von Eye, A. (2005). Positive youth development, participation in community youth development programs, and community contributions of fifth‐grade adolescents: Findings from the first wave of the 4‐H study of positive youth development. Journal of Early Adolescence, 25(1), 17–71. 10.1177/0272431604272461 [DOI] [Google Scholar]

[aphw70211-bib-0027] Lerner, R. M. , Lerner, J. V. , Almerigi, J. , & Theokas, C. (2006). Dynamics of individual ←→ context relations in human development: A developmental systems perspective. In Comprehensive handbook of personality and psychopathology, vol. 1: personality and everyday functioning (pp. 23–43). John Wiley & Sons, Inc. [Google Scholar]

[aphw70211-bib-0029] Lerner, R. M. , Lerner, J. V. , Bowers, E. P. , & Geldhof, G. J. (2015). Positive youth development and relational‐developmental‐systems. In Handbook of child psychology and developmental science (pp. 1–45). Wiley. 10.1002/9781118963418.childpsy116 [DOI] [Google Scholar]

[aphw70211-bib-0030] Lerner, R. M. , Tirrell, J. M. , Dowling, E. M. , Geldhof, G. J. , Gestsdóttir, S. , Lerner, J. V. , King, P. E. , Williams, K. , Iraheta, G. , & Sim, A. T. R. (2019). The end of the beginning: Evidence and absences studying positive youth development in a global context. Adolescent Research Review, 4(1), 1–14. 10.1007/s40894-018-0093-4 [DOI] [Google Scholar]

[aphw70211-bib-0031] Lin, L. , Hua, L. , & Li, J. (2022). Seeking pleasure or growth? The mediating role of happiness motives in the longitudinal relationship between social mobility beliefs and well‐being in college students. Personality and Individual Differences, 184, 111170. 10.1016/j.paid.2021.111170 [DOI] [Google Scholar]

[aphw70211-bib-0032] Long, Q. , Wang, Y. , Axelin, A. , Zheng, F. , Cao, Z. , Li, X. , & Guo, J. (2025). Enhancing happiness and life satisfaction in university students: Analysis with a machine learning approach. BMC Psychology, 14(1), 45. 10.1186/s40359-025-03809-3 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0033] Lundberg, S. , & Lee, S.‐I. (2017). A unified approach to interpreting model predictions. arXiv (Cornell University). 10.48550/ARXIV.1705.07874 [DOI]

[aphw70211-bib-0034] Lundberg, S. M. , Erion, G. , Chen, H. , DeGrave, A. , Prutkin, J. M. , Nair, B. , Katz, R. , Himmelfarb, J. , Bansal, N. , & Lee, S.‐I. (2020). From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2(1), 56–67. 10.1038/s42256-019-0138-9 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0035] Marcoen, A. , Goossens, L. , & Caes, P. (1987). Lonelines in pre‐through late adolescence: Exploring the contributions of a multidimensional approach. Journal of Youth and Adolescence, 16(6), 561–577. 10.1007/BF02138821 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0036] Martin‐Barrado, A. D. , & Gomez‐Baya, D. (2024). A scoping review of the research evidence of the developmental assets model in Europe. Frontiers in Psychology, 15, 1407338. 10.3389/fpsyg.2024.1407338 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0037] Masten, A. S. , Lucke, C. M. , Nelson, K. M. , & Stallworthy, I. C. (2021). Resilience in development and psychopathology: Multisystem perspectives. Annual Review of Clinical Psychology, 17(1), 521–549. 10.1146/annurev-clinpsy-081219-120307 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0038] Meng, H. , He, S. , Guo, J. , Wang, H. , & Tang, X. (2025). Applying machine learning to understand the role of social–emotional skills on subjective well‐being and physical health. Applied Psychology: HEALTH and Well‐Being, 17(1), e12624. 10.1111/aphw.12624 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0039] Milot Travers, A. S. , & Mahalik, J. R. (2021). Positive youth development as a protective factor for adolescents at risk for depression and alcohol use. Applied Developmental Science, 25(4), 322–331. 10.1080/10888691.2019.1634569 [DOI] [Google Scholar]

[aphw70211-bib-0040] Nie, W. , Gao, L. , & Cui, K. (2022). Bullying victimization and mental health among migrant children in urban China: A moderated mediation model of school belonging and resilience. International Journal of Environmental Research and Public Health, 19(12), 7135. 10.3390/ijerph19127135 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0041] Oberle, E. (2018). Early adolescents' emotional well‐being in the classroom: The role of personal and contextual assets. Journal of School Health, 88(2), 101–111. 10.1111/josh.12585 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0042] Pelt, D. H. M. , Habets, P. C. , Vinkers, C. H. , Ligthart, L. , Van Beijsterveldt, C. E. M. , Pool, R. , & Bartels, M. (2024). Building machine learning prediction models for well‐being using predictors from the exposome and genome in a population cohort. Nature Mental Health, 2(10), 1217–1230. 10.1038/s44220-024-00294-2 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0043] Ponce‐Bobadilla, A. V. , Schmitt, V. , Maier, C. S. , Mensing, S. , & Stodtmann, S. (2024). Practical guide to SHAP analysis: Explaining supervised machine learning model predictions in drug development. Clinical and Translational Science, 17(11), e70056. 10.1111/cts.70056 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0044] Prokhorenkova, L. , Gusev, G. , Vorobev, A. , Dorogush, A. V. , & Gulin, A. (2017). CatBoost: Unbiased boosting with categorical features. arXiv (Cornell University), 31, 6639–6649. 10.48550/ARXIV.1706.09516 [DOI] [Google Scholar]

[aphw70211-bib-0045] Pujadas, E. R. , Díaz‐Caneja, C. M. , Stevanovic, D. , Quintero, M. F. , Martín‐Isla, C. , Hernández‐González, J. , Atehortúa, A. , Lazrak, N. , Pries, L. , Delespaul, P. , Camacho, M. , Gülöksüz, S. , Rutten, B. P. F. , & Lekadir, K. (2025). Longitudinal prediction of mental health outcomes in vulnerable youth using machine learning. Cognitive Computation, 17(5), 152. 10.1007/s12559-025-10509-y [DOI] [Google Scholar]

[aphw70211-bib-0046] Radloff, L. S. (1977). The CES‐D scale: A self‐report depression scale for research in the general population. Applied Psychological Measurement, 1(3), 385–401. 10.1177/014662167700100306 [DOI] [Google Scholar]

[aphw70211-bib-0047] Roth, J. L. , & Brooks‐Gunn, J. (2003). What exactly is a youth development program? Answers from research and practice. Applied Developmental Science, 7(2), 94–111. 10.1207/S1532480XADS0702_6 [DOI] [Google Scholar]

[aphw70211-bib-0048] Salih, A. M. , Raisi‐Estabragh, Z. , Galazzo, I. B. , Radeva, P. , Petersen, S. E. , Lekadir, K. , & Menegaz, G. (2025). A perspective on explainable artificial intelligence methods: SHAP and LIME. Advanced Intelligent Systems, 7(1), 2400304. 10.1002/aisy.202400304 [DOI] [Google Scholar]

[aphw70211-bib-0049] Scott, S. , Briskman, J. , Woolgar, M. , Humayun, S. , & O'Connor, T. G. (2011). Attachment in adolescence: Overlap with parenting and unique prediction of behavioural adjustment. Journal of Child Psychology and Psychiatry, 52(10), 1052–1062. 10.1111/j.1469-7610.2011.02453.x [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0051] Shek, D. T. L. , Yu, L. , & Fu, X. (2013). Confucian virtues and Chinese adolescent development: A conceptual review. Ijamh, 25(4), 335–344. 10.1515/ijamh-2013-0031 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0050] Shek, D. T. , Zhao, L. , Dou, D. , Zhu, X. , & Xiao, C. (2021). The impact of positive youth development attributes on posttraumatic stress disorder symptoms among Chinese adolescents under COVID‐19. Journal of Adolescent Health, 68(4), 676–682. 10.1016/j.jadohealth.2021.01.011 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0052] Stekhoven, D. J. , & Bühlmann, P. (2012). MissForest—Non‐parametric missing value imputation for mixed‐type data. Bioinformatics, 28(1), 112–118. 10.1093/bioinformatics/btr597 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0053] The Lancet Psychiatry . (2025). New perspectives on prevention. The Lancet Psychiatry, 12(10), 723. 10.1016/S2215-0366(25)00281-0 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0054] Wang, Y. , Su, S. , Xiao, J. , & Lin, D. (2025). Positive character and psychological well‐being: A nuanced examination among adolescents from diverse rural–urban and migration contexts. Journal of Adolescence, 97(1), 180–195. 10.1002/jad.12409 [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0055] Wen, M. , & Lin, D. (2012). Child development in rural China: Children left behind by their migrant parents and children of nonmigrant families. Child Development, 83(1), 120–136. 10.1111/j.1467-8624.2011.01698.x [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0056] Wen, M. , Su, S. , Li, X. , & Lin, D. (2015). Positive youth development in rural China: The role of parental migration. Social Science and Medicine, 132, 261–269. 10.1016/j.socscimed.2014.07.051 [DOI] [PMC free article] [PubMed] [Google Scholar]

[aphw70211-bib-0057] World Health Organization . (2024). The adolescent health indicators recommended by the global action for measurement of adolescent health. World Health Organization. https://books.google.com/books?id=A6cOEQAAQBAJ [Google Scholar]

[aphw70211-bib-0058] Xiang, S. , Zhao, S. , Xiao, J. , Wang, Y. , Li, Y. , Liu, Z. , & Lin, D. (2024). Longitudinal trajectories of social mobility beliefs among Chinese adolescents: The protective roles of parental academic involvement and adolescent future orientation. Journal of Youth and Adolescence, 53(8), 1903–1917. 10.1007/s10964-024-01984-x [DOI] [PubMed] [Google Scholar]

[aphw70211-bib-0059] Yuan, R. , & Liu, M. (2025). Promoting positive growth through building developmental assets for maltreated adolescents: A multiple mediation analysis. Child Abuse Review, 34(3), e70044. 10.1002/car.70044 [DOI] [Google Scholar]

PERMALINK

Predicting positive youth development among Chinese adolescents: A machine learning approach using multiwave longitudinal data

Zelin Liu

Zékai Lu

Yaqiong Wang

Danhua Lin

Abstract

INTRODUCTION

The Present Study

FIGURE 1.

DATA AND METHODS

Participants and Data

TABLE 1.

Measures

Outcome Variable

Predictor variables

Demographics and personality traits

Internal psychological resources

Family and parental factors

School and community context

Risk factors and health behaviors

Missing Data

ANALYTIC STRATEGY

RESULTS

Model Comparison and Selection

TABLE 2.

Global Feature Importance

FIGURE 2.

Age Group Heterogeneity

FIGURE 3.

FIGURE 4.

Student Type Heterogeneity

FIGURE 5.

DISCUSSION

School Psychological Climate as the Core External Resource

Depression and Parental Loneliness as Key Risk Factors

Social Mobility Belief and Resilience as Internal Developmental Assets

Age‐Graded Developmental Transition

Student Type Heterogeneity and Three Developmental Pathways

Theoretical and Practical Implications

Limitations and Future Directions

CONCLUSION

CONFLICT OF INTEREST STATEMENT

ETHICS STATEMENT

Supporting information

ACKNOWLEDGMENTS

DATA AVAILABILITY STATEMENT

REFERENCES

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases