Correlates of unsafe driving in non-professional drivers: insights from a data-driven modeling approach

Abstract

Background

Aberrant driving behavior is a major determinant of traffic crashes and injuries. It encompasses deliberate violations, errors, and lapses. Understanding the specific driver characteristics associated with each behavioral domain is critical for improving road safety. This study aimed to identify demographic, exposure-related, and behavioral covariates associated with violations, errors, and lapses as measured by the Driver Behavior Questionnaire.

Methods

A secondary analysis was conducted using a population-based dataset of 1,764 non-professional drivers from Egypt. Drivers completed questionnaires assessing socio-demographic factors, driving exposure, and habitual behaviors. Three separate penalized linear regression models using elastic net regularization were developed for violations, errors, and lapses total scores. Model tuning was optimized by repeated 10-fold cross-validation, followed by refitting selected predictors.

Results

Low education, mobile phone use, eating while driving, and driving while sleepy were independently associated with higher violation scores (p < 0.001). Errors were significantly associated with increasing age, female sex, low education, non-use of seatbelts, eating while driving, and sleepiness (p ≤ 0.005). Lapses were predicted by increasing age, female sex, low education, non-use of seatbelts, and driving while sleepy (p ≤ 0.001). Driving experience was inversely associated with all aberrant behavior domains (p < 0.001). Model diagnostics confirmed adequate fit, low multicollinearity, and robust predictive performance.

Conclusions

Distinct behavioral, demographic, and exposure-related factors predict violations, errors, and lapses among non-professional drivers. These findings underscore the multifactorial nature of aberrant driving and support the development of behavior-specific, evidence-based interventions to enhance road safety and inform precision-focused management of driver behavior strategies.

Highlights

Findings reveal heterogeneous mechanisms underlying unsafe driving behaviors.

Educational level and risky habits were key predictors of violations and errors.

Sleepiness, mobile phone use, and eating while driving increased aberrant behavior.

Age and sex showed differential effects across errors and lapses models.

Results support targeted, behavior-specific driver safety interventions.

Introduction

Aberrant driving behavior is a dominant factor in road traffic crashes, significantly contributing to global injury and fatality burdens [1]. It reflects core challenges for traffic safety research and policy interventions [2]. Human behavioral risk factors, such as speeding, distraction, and deliberate non-compliance, are identified as primary contributors to crash incidence [3]. Recent studies underline the multifaceted nature of driver behavior determinants [4]. Socio-demographic predictors, such as age and sex, have been suggested to be associated with scores on violations and lapses [5]. Driving experience and educational attainment tend to moderate errors and lapses across observational studies [5]. Self-reported aberrant driving measures such as the Driver Behaviour Questionnaire (DBQ) capture distinct dimensions of violations, errors, and lapses that demonstrate stability and external relevance in diverse driver populations [5]. Empirical analyses have shown that DBQ subscales correlate with real-world risky driving outcomes [1]. However, integrated models are needed to identify which driver characteristics systematically relate to violations, errors, and lapses [1, 6].

Knowledge gaps exist in understanding the multivariate role of driver characteristics in aberrant driving behavior [7]. Evidence suggests that demographic factors alone are insufficient to predict DBQ subscale scores [8]. This limits the generalizability of existing models linking violations, errors, and lapses to individual traits and crash risk [3]. Some studies identified factors affecting lapses and violations; however, they do not test their incremental contributions to each DBQ outcome [8]. This leaves a gap in modeling complex covariate structures essential for targeted interventions [9]. The relative contribution of driver characteristics and behavioral styles to the distinct DBQ subscales remains under-examined across broad adult driver populations rather than specific subgroups such as young or professional drivers only [4, 10, 11]. In addition, few studies have simultaneously investigated the covariate patterns unique to each DBQ dimension (violations, errors, lapses) using robust statistical models [3]. There is ambiguity about whether specific traits differentially predict intentional versus unintentional aberrant behaviors [3]. Research also lacks clarity on how driver experience interacts with other factors, such as sex, to shape aberrant behavior profiles, making it difficult to tailor behavioral safety interventions effectively [4]. Addressing these gaps is essential for advancing theoretical understanding of aberrant driving behaviors and for developing predictive models that can inform targeted road safety strategies.

We aimed to identify driver characteristics independently associated with violations, errors, and lapses, as measured by the DBQ, using multivariable regression modeling. We hypothesized that distinct demographic, exposure, and behavioral factors would differentially predict each DBQ subscale, reflecting heterogeneous underlying mechanisms of intentional and unintentional aberrant driving behaviors. This investigation was necessary to provide a predictive hierarchy of fixed or modifiable driver characteristics influencing specific behavioral dimensions. The findings hold implications for transportation safety, as they enable the development of data-driven models for identifying high-risk driver subgroups.

Materials and methods

Study population

We conducted a secondary analysis of data from a population-based cross-sectional study conducted on non-professional drivers in Beni-Suef, Egypt, from October to December 2019. In a previous study, Arafa et al. investigated age-related differences in driving behaviors among non-professional drivers. Non-professional drivers were defined as individuals who drove their own vehicle or one owned by friends or relatives and were not employed as part-time or full-time drivers. To be eligible, they had to be at least 19 years old and have a minimum of one year of driving experience. To provide a representative sample, they used a stratified random sampling. One area was randomly selected from each stratum, and eligible drivers were recruited through door-to-door household visits using a random start method. Participants completed a questionnaire consisting of two sections. The first section collected demographic information, driving exposure, and habits, including seatbelt non-use, mobile phone use, eating while driving, and driving while sleepy or drowsy. The second part included the DBQ questionnaire. In this study, we used the dataset provided by Arafa et al. to develop three models for each subscale of the DBQ scores.

Data availability

The anonymized dataset is publicly accessible at (https://doi.org/10.1371/journal.pone.0238516.s001), as a supplementary file (doi: https://doi.org/10.1371/journal.pone.0238516; last access date: January 5, 2026) to Arafa’s open-access article [12] distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Ethical considerations

The Arafa’s study protocols were approved by the Research Ethics Committee of the Faculty of Medicine, Beni-Suef University [12].The study was conducted in accordance with the Declaration of Helsinki. Signed informed consent was obtained from participants. In our study, we did not perform any direct measurements or access to personally identifiable information. Instead, we conducted a secondary analysis using fully de-identified data from the publicly available database. The original information source and dataset were appropriately acknowledged, and no raw data were reproduced or republished.

Study variables

We included age (years), sex, educational level (elementary or high), driving hours per day, driving experience (years), and driving habits, including eating, using a mobile phone while driving, not using a seatbelt, and driving while feeling sleepy. The ordered categories were measured on a four-level scale (never, seldom, sometimes, and always) [12]. The DBQ questions were scaled from zero to five (never, hardly ever, occasionally, quite often, frequently, and nearly all the time) [13]. The questionnaire is a widely validated instrument and has demonstrated robust psychometric performance across diverse populations [14]. It was used for quantifying dimensions of aberrant driving behaviors. Three aberrant driving behaviors, including violations, errors, and lapses, were recorded for each driver. The items were divided as follows: ten items for violations such as “disregarding the speed limits on a motorway”, eight items for errors such as “missing signs”, and eight items for lapses such as “getting into wrong lanes.” High scores of DBQ indicated aberrant driving behavior.

Data analysis

We developed three models for each subscale of the DBQ scores adjusted for drivers’ characteristics. Exploratory data analysis was performed to characterize variable distributions. Descriptive summaries and graphical diagnostics were used to identify potential anomalies and inform subsequent modeling decisions. To address potential redundancy and high-dimensional correlation structures among predictors, feature selection was embedded within the modeling process. In preliminary regression models, we used Cook’s distance to find influential observations for each outcome-specific (violations, errors, and lapses) linear regression model. Two items within the DBQ violations subscale pertain to mobile phone use and seatbelt non-use. We decided to include the two as independent covariates in the violations model. This decision reflects the conceptual distinction between the outcome and predictors. The violations score represents a composite latent construct aggregating multiple heterogeneous intentional driving behaviors, whereas mobile phone use and seatbelt non-use were operationalized as externally defined, ordinal indicators of habitual driving practices capturing structured exposure gradients beyond item-level responses. The analytical objective was to evaluate whether these behaviors exhibit incremental explanatory value for overall violation on the full covariate set, rather than to model specific DBQ items. Potential redundancy and collinearity were explicitly addressed through elastic-net regularization with cross-validated tuning, which penalizes non-informative or overlapping predictors. This was supported further by low variance inflation factors. Accordingly, inclusion of these covariates constituted a theoretically grounded and statistically controlled approach to isolating behavior-specific associations within a composite outcome framework. Inferential analyses were based on penalized linear regression models using elastic-net regularization, which combines ℓ₁ and ℓ₂ penalties to simultaneously shrink coefficients and variable selection while mitigating multicollinearity. Separate models were specified for each outcome. Hyperparameters balancing between penalties and the degree of regularization were optimized via repeated 10-fold cross-validation, minimizing prediction error. Model performance was evaluated using cross-validated metrics appropriate for continuous outcomes, including mean squared error, root mean squared error, and variance-explained index adjusted R². To facilitate statistical inference, predictors retained by the penalized models were subsequently refit using ordinary least squares on the reduced feature sets, yielding unbiased coefficient estimates, standard errors, and corresponding hypothesis tests. Model diagnostics included assessment of multicollinearity using variance inflation factors, ensuring that retained predictors exhibited acceptable independence. Residual diagnostics were conducted through graphical evaluation of fitted-versus-residual relationships, including nonparametric smoothing to assess systematic deviations from modeling assumptions. Results are presented as median [interquartile range] for continuous variables, and as absolute numbers (%) for categorical data. The level of significance was set at two-tailed α = 0.05. Data analyses and visualization were performed with R software version 4.0.2. (R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/). Visualization and figure assembly were performed using ggplot2 and related extensions to ensure publication-quality outputs.

Results

Sample and preprocessing

The sample included 1764 drivers. There were no missing data or duplicate rows in the dataset. The primary analyses were conducted on the full cohort. The description of the whole sample is presented in Table 1. Figure 1 illustrates the distribution of total scores for violations, errors, and lapses. The scores of the three DBQ components are skewed to the right. This shows that most participants are ordinary drivers with low total scores of violations, errors, and lapses. Influential observations were evaluated separately for each linear regression model using Cook’s distance. For each model, observations exceeding the conventional influence threshold of 4/n were identified and excluded in outcome-specific analyses. This procedure resulted in the exclusion of approximately 5.2% of observations for the violations model, 6.5% for the errors model, and 5.7% for the lapses model. All exclusions were performed to improve the robustness of the findings.

Table 1.

Characteristics of the study sample

Characteristic		Value
N		1764
Age (year)		34.0 [23.0, 52.0]
Male		1364 (77.3)
Married		1156 (65.5)
Have Child		1053 (59.7)
High Education		1027 (58.2)
Mobile Use	never	412 (23.4)
	seldom	774 (43.9)
	sometimes	384 (21.8)
	always	194 (11.0)
Eating while Driving	never	614 (34.8)
	seldom	733 (41.6)
	sometimes	319 (18.1)
	always	98 (5.6)
Seatbelt Non-use	never	305 (17.3)
	seldom	644 (36.5)
	sometimes	418 (23.7)
	always	397 (22.5)
Sleepy	never	1167 (66.2)
	seldom	430 (24.4)
	sometimes	135 (7.7)
	always	32 (1.8)
Daily Driving (hours)		4.0 [2.0, 6.0]
Driving History (year)		7.0 [3.0, 17.0]
Total Violations (score)		14.0 [7.0, 22.0]
Total Errors (score)		12.0 [8.0, 17.0]
Total Lapses (score)		11.0 [7.0, 17.0]

Continuous variables are presented as median [interquartile range], and categorical variables as count (%)

Fig. 1.

Boxplot distributions of DBQ component scores in the total sample. Most participants were typical drivers, recording low overall scores for violations, errors, and lapses

Development of models

Hyperparameter optimization was conducted using 10-fold cross-validated elastic-net regression for each outcome. For the violations model, the optimal elastic-net specification corresponded to an intermediate mixing parameter (α = 0.30) with a regularization strength of λ = 0.151, yielding a minimum cross-validated root mean squared error (RMSE) of 6.904 and R-squared (RS) of 0.481. The errors model favored a near-lasso solution (α = 0.90) with a comparatively weak penalty (λ = 0.008), achieving the lowest RMSE of 4.962 and RS of 0.334. For the lapses, the optimal model reduced to pure ridge regression (α = 0.00) with λ = 0.197, yielding an RMSE of 28.69 and an RS of 0.393. Variance inflation factors were computed for all predictors selected by the optimally tuned elastic-net models to evaluate multicollinearity. Across all models, variance inflation values ranged from 1.033 to 3.670, indicating low multicollinearity and suggesting that the regression coefficients can be interpreted reliably without concern for inflated variance due to predictor intercorrelation. Figure 2 illustrates the plots of residuals versus fitted values for the three models. It shows that there was no evidence of systematic patterns or curvature, with residuals appearing randomly distributed around zero across the range of fitted values. In addition, locally weighted smoothing curves were approximately horizontal and centered near the zero line, supporting the adequacy of the linear functional form assumed in the regression models. Collectively, these findings indicate that outcome-specific regularization structures are present across the behavioral domains captured by the three models.

Fig. 2.

Plots of residuals versus fitted values for violations (A), errors (B), and lapses (C) model. Residuals showed no systematic patterns, appearing randomly scattered around zero. The smoothing curves were nearly horizontal and centered on zero, confirming the adequacy of the regression models

Models’ covariates

Table 2 presents the results of the three models. In the penalized regression analyses, several candidate predictors were not retained in the final models, as their corresponding coefficients were shrunk to zero at the optimal cross-validated elastic-net tuning parameters. This behavior reflects the regularization mechanism of elastic-net regression, which simultaneously performs coefficient shrinkage and automatic variable selection by balancing ℓ1 and ℓ2 penalties. Variables, such as age in the violations model, were excluded because, conditional on the full multivariable feature space and accounting for collinearity and shared variance among predictors, they did not contribute meaningful incremental predictive information for the outcome. Accordingly, only predictors with non-zero penalized coefficients were carried forward into the post-selection ordinary least squares refitting stage for statistical inference. The resulting models therefore represent stability-optimized representations of the data, emphasizing predictors with robust independent associations while mitigating overfitting and multicollinearity. As presented in Table 2, the violations model retained a subset of predictors demonstrating statistically independent associations after multivariable adjustment, including educational attainment, higher-frequency categories of mobile phone use, eating while driving, sleepiness, daily driving exposure, and driving history, and demographic indicators such as marital status and parenthood were not associated with the outcome. The errors model exhibited a broader pattern of statistically significant associations, with age, sex, educational level, higher levels of eating while driving, seatbelt non-use, sleepiness, daily driving exposure, and driving history contributing independently to model fit. In contrast, the lapses model was characterized by significant associations with age, sex, educational attainment, eating while driving, non-use of seatbelt, sleepiness, and driving history, while daily driving exposure was not retained. Collectively, these results demonstrate distinct covariate relevance profiles across the three outcome domains, supporting differential statistical determinants of violations, errors, and lapses within the penalized multivariable modeling framework.

Table 2.

Model specifications for violations, errors, and lapses

Variable		Violations Model		Errors Model		Lapses Model
		Coeff. (std)	p	Coeff. (std)	p	Coeff. (std)	p
Intercept		7.601 (0.704)	< 0.001*	8.940 (0.612)	< 0.001*	8.835 (0.657)	< 0.001*
Age		-	-	0.056 (0.014)	< 0.001*	0.060 (0.015)	< 0.001*
Male		0.601 (0.447)	0.179	-1.494 (0.324)	< 0.001*	-2.071 (0.345)	< 0.001*
Married		-0.032 (0.659)	0.961	-0.660 (0.496)	0.183	0.329 (0.534)	0.537
Have Child		-0.161 (0.643)	0.802	0.613 (0.479)	0.201	-0.352 (0.515)	0.495
High Education		-1.170 (0.376)	0.002*	-1.260 (0.275)	< 0.001*	-0.987 (0.297)	0.001*
Mobile Use	never	Reference
	seldom	0.604 (0.465)	0.195	0.199 (0.339)	0.558	-0.355 (0.365)	0.331
	sometimes	4.118 (0.576)	< 0.001*	1.733 (0.421)	< 0.001*	1.084 (0.451)	0.434
	always	3.256 (0.682)	< 0.001*	0.680 (0.493)	0.168	-0.412 (0.526)	0.433
Eating while Driving	never	Reference
	seldom	2.712 (0.422)	< 0.001*	1.058 (0.309)	< 0.001*	1.518 (0.333)	< 0.001*
	sometimes	5.507 (0.559)	< 0.001*	3.131 (0.411)	< 0.001*	3.529 (0.440)	< 0.001*
	always	6.995 (0.886)	< 0.001*	2.813 (0.639)	< 0.001*	3.249 (0.679)	< 0.001*
Seatbelt Non-use	never	Reference
	seldom	-		1.201 (0.363)	0.001*	0.964 (0.394)	0.015*
	sometimes	0.493 (0.415)	0.235	1.750 (0.393)	< 0.001*	1.850 (0.426)	< 0.001*
	always	0.367 (0.457)	0.421	1.966 (0.410)	< 0.001*	2.055 (0.440)	< 0.001*
Sleepy	never	Reference
	seldom	4.116 (0.426)	< 0.001*	2.008 (0.307)	< 0.001*	2.760 (0.330)	< 0.001*
	sometimes	6.876 (0.688)	< 0.001*	4.328 (0.523)	< 0.001*	6.602 (0.556)	< 0.001*
	always	6.846 (1.431)	< 0.001*	4.334 (1.103)	< 0.001*	8.726 (1.324)	< 0.001*
Daily Driving		0.606 (0.079)	< 0.001*	0.164 (0.058)	0.005*	0.012 (0.063)	0.855
Driving History		-0.107 (0.018)	< 0.001*	-0.118 (0.019)	< 0.001*	-0.121 (0.020)	< 0.001*

Discussion

We sought to identify driver characteristics independently associated with the three sub-dimensions of aberrant driving behavior, violations, errors, and lapses, in a large sample of non-proficient drivers. The findings demonstrated that distinct sets of demographic, exposure, and habitual factors contributed differentially to each behavioral dimension, underscoring the multifactorial and domain-specific nature of aberrant driving. Overall, educational attainment and habitual risky practices, such as mobile phone use, eating while driving, and driving while sleepy, emerged as salient covariates, whereas the influence of sex, age, and driving experience varied across behavioral domains. These results extend prior knowledge by empirically implying that intentional and unintentional driving deviations are not governed by uniform determinants but rather by distinct behavioral mechanisms reflecting cognitive control, risk perception, and self-regulatory capacity. Collectively, this study is new in simultaneously modeling violations, errors, and lapses to identify distinct and independent covariate structures within a non-proficient driver population. By integrating behavioral, demographic, and exposure variables, it isolates stable predictors while controlling for shared variance across behavioral domains. This multivariate approach improved precision in risk profiling and deepened understanding of aberrant driving mechanisms. The use of penalized regression modeling provided methodological rigor by addressing collinearity and isolating stable independent associations, offering a clear advantage over traditional multivariate techniques. Specifically, elastic-net regularization enhanced model robustness and generalizability by minimizing overfitting and effectively controlling multicollinearity. These methodological features ensure that the findings reliably inform evidence-based, behavior-specific safety interventions targeting modifiable risks.

Our findings align with evidence emphasizing the multidimensional structure of aberrant driving behavior. The association of low educational attainment and risky habits with higher violation scores concurs with research in China and India, where education level inversely predicted violations [8, 15]. The observed influence of fatigue and sleepiness on violations and lapses corroborates physiological studies linking driver fatigue to reduced vigilance and impaired motor coordination [16]. The relationship between multitasking behaviors and DBQ subscales is compatible with the evidence that mobile phone use induces cognitive load and executive interference during driving [17]. Consistent with our finding that age predicted errors and lapses, studies in Lebanon and Qatar demonstrated higher attentional failures among younger and middle-aged males due to heightened risk-taking propensity [18, 19]. The protective effect of driving experience on all DBQ subscales mirrors findings among Pakistani motorcyclists transitioning to car driving, where accumulated procedural knowledge mitigated risky acts [20]. Similarly, studies in Thailand and Ecuador expanded the DBQ framework to professional drivers, confirming fatigue and technology use as major determinants of violations and errors [21, 22]. Our results also parallel evidence from a Greek study, where impulsivity and low risk awareness were found to be central to repeated violations [23]. Collectively, our findings demonstrated that even among non-professional drivers, habitual risky practices are associated with distinct forms of aberrant driving.

The distinct predictors identified across the three behavioral domains suggest differentiated neurocognitive and psychosocial pathways underlying aberrant driving. Violations, driven by factors such as low educational attainment, frequent mobile phone use, eating while driving, sleepiness, and longer driving exposure, arise from risk-taking modulated by impaired inhibitory control and distorted risk appraisal [24]. These behaviors reflect intentional deviations associated with sensation-seeking and executive dysfunction within brain prefrontal regulatory circuits, particularly when cognitive load is increased by multitasking or fatigue [2]. Frequent mobile use during driving activates dual-task interference mechanisms, reducing visuospatial attention and increasing impulsive responding [10, 25]. Educational attainment may exert a protective influence through enhanced hazard perception and metacognitive self-monitoring [11, 26]. Sleepiness-related violations suggest a neurobehavioral overlap with microsleep-induced attentional disengagement and decreased cortical arousal [1]. Errors were more influenced by age, sex, education, multitasking behaviors, seatbelt non-use, and fatigue. These reflect unintentional failures in perceptual encoding and decision integration associated with age-related decline in attentional flexibility and slower cognitive processing speed [3]. Men showed fewer errors, possibly due to differential self-efficacy beliefs or risk habituation in complex traffic environments [4]. Eating while driving likely generates sensorimotor competition for attentional resources, disrupting working memory consolidation during routine maneuvers [8]. Sleep-related fatigue induces cortical hypoactivation within fronto-parietal attention networks, facilitating lapses in situational awareness that manifest as unintentional errors [12]. Lapses, predicted by age, sex, education, eating while driving, seatbelt non-use, sleepiness, and driving history, represent disruptions in automaticity and memory retrieval during habitual driving [3]. Their age dependence reflects the degradation of working memory and prospective recall mechanisms, often described as executive-memory dissociation in routine cognitive tasks [14]. The negative association with driving experience supports a compensatory adaptation hypothesis, whereby learned procedural control mitigates attentional slips [7]. The persistence of fatigue and secondary-task interference across both lapses and errors suggests shared cognitive resource depletion mechanisms mediated by the thalamo-cortical vigilance system [13]. Collectively, these mechanisms indicate that violations are rooted in intentional control failure, whereas errors and lapses stem from cognitive-executive degradation, attentional fatigue, and impaired working memory integration.

The present study provided actionable insights for designing targeted interventions to improve driving safety. By identifying distinct driver characteristics associated with violations, errors, and lapses, the findings support the development of behavior-specific strategies, in addition to generic road safety campaigns. Targeted interventions should prioritize reducing mobile phone use, eating while driving, and driving while sleepy, as these behaviors consistently predicted higher violations, errors, and lapses. Enforcement campaigns should integrate technology-based monitoring, such as in-vehicle mobile use detection and fatigue alert systems, to mitigate deliberate and fatigue-related transgressions. Programs enhancing seatbelt compliance and discouraging multitasking should be prioritized, particularly among younger and less-educated drivers, who demonstrated greater susceptibility to risky violations. Sex differences also should be considered in designing training programs (and further research studies). Educational initiatives must strengthen hazard perception and cognitive control training to address errors linked with low educational attainment and prolonged driving exposure. Finally, integrating fatigue management modules and behavioral feedback systems into licensing and re-certification processes could reduce attentional lapses among drivers with limited driving experience, reinforcing long-term behavioral safety adaptation. In public health and transportation management contexts, these findings support precision-based safety policies that align behavioral risk assessment to enhance the effectiveness of road safety interventions.

Limitations

This study has several limitations that should be considered when interpreting the findings. First, its cross-sectional design precludes causal inference, as associations between driver characteristics and aberrant behavior cannot establish temporal directionality. However, the analytical framework used enhances internal validity by mitigating multicollinearity and overfitting, thus strengthening the reliability of observed associations. Second, behavioral data were derived from self-reported measures (DBQ), which may be influenced by social desirability or recall bias. Nonetheless, the DBQ is a well-validated instrument with demonstrated psychometric robustness and external relevance across diverse driver populations. Moreover, the large sample size improved the reliability of parameter estimates and minimized Type II errors. The stratified sampling design also provided a representative view of ordinary drivers within the study setting. Future research should integrate longitudinal designs and objective driving performance metrics to further delineate causal pathways and strengthen the predictive validity of aberrant driving behavior models.

Conclusion

This study identified driver characteristics associated with distinct dimensions of aberrant driving behavior (violations, errors, and lapses) using a robust penalized regression modeling framework. The findings demonstrate that intentional and unintentional deviations from safe driving are influenced by heterogeneous behavioral, demographic, and exposure-related factors, emphasizing the multifactorial nature of driver risk. By differentiating the predictors of each behavioral domain, the study advanced current understanding of how specific driver traits and habits contribute to unsafe driving patterns. These insights hold practical value for developing evidence-based, behavior-specific interventions and policies aimed at improving road safety. From a public health perspective, the results underscore the importance of targeted strategies that integrate behavioral education, fatigue management, and risk communication to mitigate preventable crashes. Overall, the study contributes to a more precise data-driven foundation for traffic safety research and the design of individualized driver behavior interventions.

Author contributions

Both authors contributed to the study’s conceptualization, literature review, planning and design, data analysis, interpretation of the results, manuscript drafting, and final approval of the submitted version.

Funding

This research is supported by the research grant of the Imam Khomeini International University (271340).

Data availability

The anonymized dataset is publicly available at (https://doi.org/10.1371/journal.pone.0238516.s001) as a supplementary file (doi: https://doi.org/10.731/journal.Pone; last access date: January 5, 2026) to Arafa’s open-access article distributed under the terms of Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Declarations

Competing interests

The authors declare no competing interests.