Sustainable training practices: predicting job satisfaction and employee behavior using machine learning techniques

Abstract

This study evaluates Sustainable Training Practices (STP) that promote organizational growth and ensure the attainment of sustainable HRM objectives. First, we employ Structural Equation Modelling to identify relationships between STP, Psychological Contract Fulfilment, Job Satisfaction, and Organizational Citizenship Behavior. Next, we build a predictive model using the RF Regression Supervised Machine Learning technique to identify the key predictors. Our findings indicate that employee happiness, expectation fulfilment, and behavior are highly dependent on the STPs offered to them. In addition, we find that machine learning is crucial because it reveals hidden features that are sometimes overlooked by conventional methods.

Introduction

In 2020, India passed it ground-breaking New Education Policy (NEP, 2020), designed to help the country ready its’ workforce to keep pace with the tremendous growth in the economy. Along with a comprehensive re-vamping of the education set-up (see Varma et al., 2021), the policy also calls for a new look at human resource systems, such as performance management and training (see, e.g., Varma et al., 2022). Indeed, training has been identified as an essential SHRM practice for sustaining long-term career development, and addressing productivity levels, and retention and turnover rates of employees (Kramar, 2022; Stankevičienė & Savanevičienė, 2018).

Accordingly, this study was designed to address and investigate sustainable training programs implemented to meet sustainable HRM objectives. Sustainable Training Practices (STP) can be defined as programs designed to ‘increase the recipients’ self-reliance in the long run, often completely eliminating the need for future support’ (Brady et al., 2013). In other words, it is the expansion of employees’ capacities, knowledge, and skills that aids them in the present and also prepares them for future challenges that cannot be predicted. Essentially, STP enables companies to create a work environment that encourages employee progress by focusing on their personal and professional requirements. Indeed, sustainable training has been identified as one of the most effective determinants of employee satisfaction and in influencing employee behaviors (Davidescu et al., 2020; Lee, 2015; Rodriguez & Walters, 2017).

Thus, this research also focuses on employee behaviors toward their organizations that are voluntary in nature (e.g., extra-role activities) yet expected of them when their levels of expectations correspond to the total fulfilment of their psychological contracts (Mazumdar et al., 2022). These extra-role behaviors are referred to as organizational citizenship behaviors (OCBs) and are separated into two components: OCB-Organization (directed toward the organization) and OCB-Individual (directed toward individuals) to minimize confusion (Williams & Anderson, 1991). The levels of employee job satisfaction, extra-role behaviors, and fulfilled psychological contracts (Fontinha et al., 2014) are a result of the levels of STPs practiced by the organizations, such as training need detection, skill level monitoring, personal and professional needs, mentoring programs, training modules tailored to individuals, etc. Previous studies have explored the links between SHRM and job satisfaction (JS), psychological contract fulfilment (PCF), and OCB, but there is a dearth of information addressing STP, despite the fact that it has always been a critical sub-dimension of SHRM (Davidescu et al., 2020; Dixon-fowler et al., 2020; Freire & Pieta, 2022; Susomrith, 2020). Further, there is a lack of studies undertaken in the Asian context, creating a significant gap.

The present study

This paper is divided into two sections: In the first section, the study evaluates the links between job satisfaction (JS), organizational citizenship behaviors (OCB), and the fulfillment of psychological contracts by employees with their continuous sustainable training activities. In the second section, the paper develops a prediction model for forecasting JS, OCB, and PCF using STP, thereby defining objectives (a) to understand the impact of STP on employee expectations, satisfaction, and behaviors, (b) to highlight the importance of machine learning techniques in understanding these relationships in an Asian context, and (c) to develop a model for a prediction that can be used in the future for the assessment of STP.

Theoretical background

Social exchange theory (SET)

According to social exchange theory (Blau, 1960; Homans, 1958), parties to an exchange often examine each other’s behavior before engaging in an exchange based on the benefits they have gained. In other words, social exchange is generally based on calculating the cost–benefit analysis of entering into a relationship. In the workplace, the employee-employer relationship is dependent on entering and maintaining the exchange when both parties feel that their actions are mutually beneficial (Cook, 1987). More specifically, employees tend to observe the actions of their employer towards them such that when they feel that the employer is investing in them, they tend to reflect it in their actions by enhancing their performance levels (Cook et al., 2013). These transactions are not typical economic transactions, but humans anticipate intangible rewards such as employee satisfaction, respect, and proactive job practices among others. Yin (2017) demonstrated that when employees perceive that their employer has created an active job engagement environment, they tend to engage in activities that directly or indirectly benefit the organization, such as enhanced engagement in OCBs and improved task performance, which reduces burnouts. In a recent assessment of social SET was identified as the medium of these exchange relationships (Chernyak-Hai & Rabenu, 2018). Thus, exchange relationships demonstrate their strength when they are reciprocated over time and assist organizations in maintaining a balance between employee-employer relationships that can be crucial during a pandemic (Sungu et al., 2019). As a result, the emphasis of this study is on the quality of the connections created between the adoptions of STP and the production of reciprocal behavior from the organization’s personnel.

Sustainable training practices (STP) and job satisfaction (JS)

Training as a Sustainable HRM practice has taken a long-term perspective on the organization’s development into consideration (Nuninger et al., 2016; Kwon, 2019; Piwowar-Sulej, 2021). In this connection, Hansson (2007) and Hirsig et al. (2014) highlighted the beneficial outcomes of STP by conducting a cost–benefit analysis of investing in training rather than infrastructure. Their results demonstrate that training the current workforce for diverse activities is significantly more advantageous than rapidly hiring new employees (Stankevičienė & Stankevičienė, 2018). Indeed, training at an individual level affects the employee’s satisfaction levels in the organization as the employees feel that the organization is taking care of them (Sabuhari et al., 2020; Terera & Ngirande, 2014). Therefore, the quality of training programs will determine employee satisfaction levels and the nature of the working relationship between employees and employers. Employees perceive training as an opportunity for advancement in organizational and career life, leading to increased satisfaction levels, improved performance, and increased intent to work with the organization (De Vos & Van der Heijden, 2017; Giovanni Mariani et al., 2013).

Next, job satisfaction has been defined as a result of how one interprets the relationship between what they want out of their work and what they think it offers or entails (Locke, 1969). Hanaysha and Tahir (2016) point out that when workers lack the knowledge necessary to do their tasks, it significantly affects their performance. STP addresses this lacuna, by identifying the need for training and development, and adequately offering it in accordance with what is needed by the individuals. These procedures aid in the long-term growth, renewal, and development of organizational human resources (Davidescu et al., 2020; Strenitzerová & Achimský, 2019). Thus, training plays an important role in affecting and motivating employee satisfaction levels fostering long-term sustainable organization (Cho & Choi, 2021; Heimerl et al., 2020). As such, we predict:

H1

STP (sustainable training practices) is (are) positively associated with JS (job satisfaction).

Sustainable training practices (STP) and psychological contract fulfillment (PCF)

The idea of sustainable organizations has taken emerged with renewed emphasis in the age of digitalization and globalization, where the focus has shifted to the requirements of the employees and ways to improve their working environment (Chinyamurindi, 2021). The dissemination of knowledge to the workforce is essential for a long-lasting business (Jilani et al., 2020), but the relationship between the employee and the employer is what matters most (Zhang et al., 2019a); however, these relationships are becoming more challenging as a result of current global issues. To maintain these kinds of interactions, a certain level of comprehension is required (Rao, 2021). Therefore, the role of expectations needs to be examined so we can further explore these mutual exchange relationships. These expectations that extend beyond the legal contract and are of an unwritten character are referred to as psychological contracts (Gouldner, 1960; Rousseau, 1989). Thus, SET explores the cost–benefit analysis of expectations, the fulfillment of these expectations, and hence the reciprocation done by both parties. The more such expectations are met, the more their psychological contracts are fulfilled. Consequently, psychological contracts have proven essential for interpreting the relationship between employees and their employer (Estreder et al., 2021). PCF refers to the execution of responsibilities for trading between the employer and the employee (Robinson & Rousseau, 1994) and leads to a higher quality employer-employee relationship (see, e.g., Risner et al., 2023). On the other hand, if an employee has been promised better skill development and training opportunities, but the organization fails to provide such developmental programs to the employees, this may lead to a psychological contract violation made by the employer resulting in a disappointed employee (Kickul, 2001; Susomrith, 2020). Every disappointed employee contributes to a disoriented employee-employer relationship. In this connection, previous studies have shown that training led to the fulfillment of psychological contracts and brought about positive results in employees such that it has helped to reduce employee exhaustion, reduced employee turnover, and brought about behavioral changes in them (Deas & Coetzee, 2020; Delobbe et al., 2016). Accordingly, we predict:

H2

STP (sustainable training practices) is (are) positively associated with PCF (psychological contract fulfillment).

Sustainable training practices (STP) and organizational citizenship behavior (OCB)

Training procedures offer workers the idea that organizations are interested in strengthening their skills and inspire them to go beyond their assigned jobs to engage in OCB (Memon et al., 2017). Steps in the welfare of employees develop an obligation towards the employers, which is backed by the SET, such that when one party goes beyond the written responsibilities, the others tend to respond similarly to maintain the mutually beneficial relationship (Emerson, 1987). Thus, when organizations invest in STP, they enhance their involvement in employee wellbeing and encourage employees to engage in activities outside of their specific responsibilities (Jaškevičiūtė et al., 2021). In other words, training creates a sense of “job influence” to the employee, leading them to engage in extra-role behaviors such as OCB and enhancing employees’ “comfortableness” and “confidence” (Snape & Redman, 2010). As such, it follows that when employees utilize the abilities acquired through these training programs, they tend to feel more obligated, which enhances their OCB levels. In this study, we examine the direct association of STP and OCBs to better understand these concepts in real-time scenarios (see Fig. 1). As such, we predict:

Fig. 1.

Theoretical model

H3

STP (sustainable training practices) is (are) positively associated with OCB (organizational citizenship behavior).

Sustainable human resource practices and human resource (HR) analytics

The pandemic has clearly demonstrated the critical role that technology played in the conduct of organizations (Naidu, 2021). The process had already begun but geared up significantly due to the dynamic environment. It demonstrated how, with the assistance of technology, businesses survived and supported their employees during these trying times. As a result of technological developments and enhancements, employees have evolved into the organization’s most valuable assets in providing a competitive advantage (Baum, 2018; Deeba, 2020; Jaiswal et al., 2021). Through the application of technology, most HR activities have been automated and cost-effective. For example, despite the critical situation, the process of talent acquisition and attraction was easily carried out thanks to the automation of the processes (HRMS software, social media like LinkedIn, and websites like naukri.com, etc.).

Our study attempts to go a step further and resolve the training challenge by applying the human analytics approach branch. Additionally, research is ongoing towards the development of additional effective measures and the deployment of more viable methods to organizations. The fact that these approaches are forward-thinking is one of the primary factors that have contributed to their widespread adoption. Contributing to this automation, the study attempted to develop a seamless way based on a predictive strategy to understand (present employee) and anticipate (future employee) the attitudes and behavior of employees who were heavily influenced by the organization’s training.

Machine learning

As a result of the digitalization era, humans have begun to generate vast quantities of data, which has led to an exponential increase in the volume of data within businesses. Finding patterns, correlations, and trends in these data has become more complex as data heterogeneity has proliferated, making standard methods of analysis for research onerous. Hence, more advanced technologies such as machine learning are required for establishing strategies and assessing trends. Machine learning is an artificial intelligence technique that analyzes data using multiple algorithms in mining operations to classify or predict crucial findings. It focuses on the development of mathematical models that are constantly trained to enhance their accuracy in order to provide better predictive results (Bishop & Nasrabadi, 2006). For instance, an algorithm can learn to forecast whether a patient would contract a disease by analyzing daily medical reports. The algorithm will improve its performance through gaining more experience by investigating these reports on a larger population (Ray, 2019). It can be applied in two ways: Supervised learning and Unsupervised learning. Supervised learning can be defined as “the machine learning task of learning a function that maps an input to an output based on example input–output pairs”. On the other hand, in Unsupervised learning, “the algorithms are left to discover their own structure in the data” (Mahesh, 2020).

Our study adopted a supervised machine learning technique on the collected data, employing Random Forest (RF) Supervised Machine Learning predictive technique for a more in-depth analysis and the identification of hidden relations in predicting the behavior of any employee (Khera & Divya, 2018). RF Regression is a branch of machine learning adapted to deal with classification, unbalanced, complex, noisy data, and outliers (Wang & Zhang, 2020) which traditional machine learning (ML) fails to handle (Raman & Pramod, 2021). The current study highlights a very complex structure of human behaviors and relationships, and to understand them in depth we needed a tool that could highlight even a smallest association, as it could be of high importance in understanding human psychology and behavior altogether and RF technique helped the study in dealing with this complexity of data.

Methodology

Participants and setting

Participants in our study were employees of companies registered with the Securities and Exchange Board of India (SEBI) (Format for business responsibility report, 2015); 230 employees participated in the study. Self-administered questionnaires were shared with the participants via a google form. Participants were asked for basic personal information such as age, gender, and industry experience, as well as current organization experience, number of earners in the family, affiliated industry, and income group. The average age of the participants was 30, with a standard deviation of 7.8 years. Sixty-four percent of participants were male, and their average age was 30.7 years. The average experience of our participants was 2.57 years.

Measures

The research construct included the following variables on 5-point Likert scales: STP, PCF, JS, and OCB. The reliability of the theoretical construct was tested using Cronbach’s alpha resulting in 0.950.

STP

was adopted from three different scales, i.e., Hurley and Hult (1998), Diaz-Carrion et al. (2018), and Barrena-Martínez et al. (2019). A sample item is “training level is defined based on commitment,” ranging from “1” very poor to “5” excellent. The reliability of STP is 0.969.

PCF

Five global PCF items were borrowed from Conway and Briner (2005). An example of an item is “organization says it will do good things for you and then gets around to doing them,” ranging from “1” not at all to “5” to a great extent. The reliability of PCF scale is 0.862.

JS

Macdonald and Maclntyre’s (1997) 10-item job satisfaction scale was adopted. A sample item is “all wages are good,” ranging from “1” as "strongly disagree” to “5” as “strongly agree. ” The reliability of JS scale is 0.928.

OCB

Henderson et al. (2020) fourteen item scale was adopted to measure OCB. An example of an item is “I helped others who have heavy workloads,” ranging from “1” as never to “5” as always. The reliability of OCB scale is 0.736.

Demographic variables

Age, gender, income group, industry experience, present organization experience, number of earners, and job title were used as control variables (see, e.g., Brinck et al., 2020).

Statistical analysis

The statistical analysis adopted in the current experiment is done in three steps i.e., EFA, SEM and finally, RF regression supervised learning technique. The purpose of each analysis is unique and important for the study. The proposed model is theoretical in nature, therefore, its plausibility was needed to be established. The significance of the first step, EFA, was dual in nature. First, EFA was necessary as the scales adopted were experimented in different cultures and to validate the item relevancy in the current research area in Asian context was required. Secondly, the construction of a predictive system requires exploratory study with multiple iterations, in accordance with the recommendations of Ozkaya (2020) and Amershi et al. (2019). In the second step, as the theoretical construct generated a temporary path connecting the components in the second step, SEM was undertaken to analyse and validate the effectiveness of the entire model. SEM enables the construction of measurable routes between measured variables and latent variables (Streiner, 2006; Tenenhaus, 2008). Thus, SEM facilitates the recognition of a “fit” model for the data, which consists of two stages, namely the measurement model and the structural model (Kelloway, 1995). A measurement model explains the measure of latent variables via exogenous or exploratory variables, whereas the second model attempts to comprehend the direct, indirect, or any relationship between the variables (Barrett, 2007; Doloi et al., 2012). In the third step, supervised learning using RF regression was utilized to construct a prediction model. As required by the study, the primary purpose for using a small dataset and doing supervised learning was to help determine individual behavior (Sánchez et al., 2020). Compared to large datasets, smaller datasets facilitate the identification of minor correlated variables that can provide crucial context for understanding the relationship between two variables. These two measures will make the study more conclusive.

Experimental design

We used an experimental design to collect data on the critical variables in our study, namely STP, PCF, JS, and OCB.

Step 1 At the first step, Exploratory factor analysis (EFA) was conducted (Kelloway, 1995; Stapleton, 1997). As previously stated, STP was evaluated on a 5-point Likert scale. Data were checked and coded to conduct exploratory factor analysis. To facilitate analysis, the STP data items were labeled with the letter “T” and the corresponding item number (1–26), as were PCF as PC1-PC5, JS as JS1-JS10, and OCB as OCB1-OCB14.

Step 2 Component-based SEM was undertaken following the identification of the dominant items. A measuring model illustrates the link between the variables STP, JS, PCF, and OCB.

Step 3 The development of a prediction model for STP, JS, OCB, and PCF using the supervised RF regression learning technique.

Results

Exploratory factor analysis (EFA)

For conducting EFA, the factors in the data set with a score of > 0.5 were listed in Table 1 (Bulut & Culha, 2010). The KMO sample adequacy for sustaining data quality measured 0.94, or near to 1, and Bartlett’s test of sphericity was significant (χ² 7395.33, df-741, sig 0.000). In four iterations (T1, T5, T16, T17, PC1, PC2, PC3, JS5, and OCB14), many determinants were eliminated to get minimum components in the rotated factor matrix. Maximum likelihood was used for extraction, whereas varimax with Kaiser Normalization was employed for rotation (Wiktorowicz, 2017).

Table 1.

Factor loading

Factor	Factor loading	Factor	Factor loading
T21	0.793	T7	0.703
T22	0.780	T8	0.694
T26	0.774	T10	0.662
T24	0.764	T9	0.652
T20	0.759	T6	0.628
T23	0.751	T12	0.724
T25	0.712	T13	0.635
T19	0.711	T15	0.619
T18	0.677	T14	0.564
T3	0.690	OCB2	0.742
T4	0.666	OCB3	0.725
T2	0.567	OCB4	0.706
PC4	0.988	OCB1	0.701
PC5	0.740	OCB5	0.671
OCB11	0.757	OCB6	0.575
OCB10	0.767	JS1	0.600
OCB12	0.687	JS2	0.624
JS3	0.769	JS4	0.569
JS6	0.599	JS7	0.637
JS8	0.646	JS9	0.646
JS10	0.801

The rotational factor matrix produced the eight most associated factors or components, namely TR1, TR2, TR3, TR4, OCBI, OCBO, JS, and PCF, as shown in Table 4. All incorporated components have the highest factor loadings. STP accounts for 50% of the variables’ variation. These indicators will serve as exploratory and latent variables in the subsequent phase of research.

Table 4.

Data set provided for predictive modelling

Factor	data item	Item description	Factor	Data item	Item description
TR1	T21	Training programs address all the needs	TR2	T7	Training level is defined on the basis of commitment
	T22	Training programs provide roadmap for professional goals		T8	Training level is defined on the basis of performance review
	T26	Training programs match the level of expectation		T10	Training level is defined on the basis of professional needs
	T24	Training programs meet your professional needs		T9	Training level is defined on the basis of personal needs
	T20	Training programs necessary to do the job well		T6	Training level is defined on the basis of talent and skill
	T23	Training programs provide roadmap for personal goals needs	TR3	T12	Organization provide opportunity for various education programs
	T25	Training programs meet your personal needs		T13	Organization encourages the employees for attending various training programs
	T19	Training programs supports to learn new skill		T15	Organization offers mentoring programs as a part of the training of the employees
	T18	Training programs are up to date		T14	Organization offers a training budget that employees had to spend themselves
TR4	T3	Working environment is continuously improving	PCF	PC4	Organization says it will do things for you and then gets around to doing them
	T4	Working environment is aspiring		PC5	I am often told I will receive things from this organization that in the end materialize
	T2	Working Environment is learning	OBI	OCB2	I helped others who have heavy workloads
JS	JS1	I receive recognition for a job well done		OCB3	I helped orient new people even though it is not required
	JS2	I feel close to the people at work		OCB4	I assisted my supervisor with his/her work (when not asked)
	JS3	I feel good about working at this company		OCB1	I helped others who have been absent
	JS4	l feel secure about my job		OCB5	I took time to listen to co-workers’ problems and worries
	JS6	On the whole, I believe work is good for my physical health		OCB6	I took a personal interest in other employees
	JS7	My wages are good	OCBO	OCB10	I took undeserved work breaks
	JS8	All my talents and skills are used at work		OCB11	A great deal of my time was spent on personal phone/email/other communications
	JS9	I get along with my supervisors		OCB12	I complained about insignificant things at work
	JS10	I feel good about my job

Path analysis and structural equation modelling (SEM)

Path analysis was conducted using SEM and AMOS software (Jayasinghe-Mudalige et al., 2012); Fig. 2 depicts the reflective measurement model and its indicator reliability measures, such as the high standard of dependability between factors. The EFA process eliminated four OCB items (OCB7, OCB8, OCB9, and OCB13) in four iterations to get the best model validity and fit measures, as shown in Tables 2, 3, and 4. This may be because OCB toward the peer did not appear to be impacted by these training programs; hence, these four components were eliminated from further analysis. Composite Reliability (CR) values of factors were set at > 0.8 and Average Value Explained (AVE) values were set at > 0.5 (Hu & Bentler, 1999), indicating convergent validity for all the constructs (Ko & Campbell, 2021). The relative CR values of the eight components were 0.968, 0.93, 0.92, 0.866, 0.888, 0.830, 0.896, and 0.863. Comparative Fit Index (CFI) > 0.95 and Standardized Root Mean Square Residual (SRMR) 0.08 (To solidify, (Root Mean Square Error of Approximation) RMSEA 0.06) are required for the best model fit measurements. After eliminating several OCB items, our model attained CFI values of 0.929, SRMR values of 0.053, and RMSEA values of 0.056, all of which were deemed acceptable and excellent by the standards (Ref. Table 3). (Kline, 2015; Gao et al., 2008).

Fig. 2.

SEM (reflective measurement model)

Table 2.

Model validity measures

	CR	AVE	MSV	MaxR(H)	1	2	3	4	5	6	7	8
1	0.968	0.774	0.587	0.970	0.879
2	0.930	0.597	0.547	0.938	0.740***	0.773
3	0.922	0.703	0.587	0.925	0.766***	0.678***	0.838
4	0.864	0.514	0.225	0.867	0.280***	0.383***	0.182*	0.717
5	0.888	0.666	0.579	0.903	0.761***	0.662***	0.691***	0.189*	0.816
6	0.830	0.620	0.225	0.834	− 0.028	− 0.087	− 0.058	− 0.475***	− 0.056	0.788
7	0.896	0.743	0.506	0.905	0.712***	0.698***	0.700***	0.215***	0.630***	− 0.076	0.862
8	0.863	0.759	0.091	0.868	− 0.064	− 0.042	− 0.052	0.032	− 0.019	0.302***	− 0.048	0.871

Significance of Correlations: ^†p < 0.100; *p < 0.050; **p < 0.010; ***p < 0.001; the bold represents the square root of average variance extracted (AVE)

Table 3.

Model fit measures

Measure	Estimate	Threshold	Interpretation
CMIN	1278.422	–	–
DF	747.000	–	–
CMIN/DF	1.711	Between 1 and 3	Excellent
CFI	0.929	> 0.95	Acceptable
SRMR	0.053	< 0.08	Excellent
RMSEA	0.056	< 0.06	Excellent
PClose	0.027	> 0.05	Acceptable

As (P 0.05), OCBI—TR1***, OCBO—TR1***, OCBO—TR3***, JS TR3***, OCBI—TR4***, OCBO—TR4***, JS—TR4***, and OCBI—TR3*** were shown to be highly statistically significant. In contrast, JS—TR1, OCBO—TR2, and JS—TR2 were 0.013, 0.011, and 0.041, respectively, and were only statistically significant. As the unidentified preferences were not recognized, the results were not statistically significant for all the correlations. That is, PCF was not shown to be significant using STP.

The findings indicate that hypothesis 1, namely STP, is positively correlated with JS. As depicted in Fig. 2, all the separate STP components have a significant connection with the JS component. Likewise, hypothesis 3, namely STP, is positively correlated with OCB. As demonstrated in Fig. 2, each component of STP is significantly associated with both components of OCB. Thus confirming that the training techniques provided to these individuals had an effect on their satisfaction and citizenship behaviors. Each component of STP was found to play a significant role in illustrating employee satisfaction and behavior. The data, however, reject hypothesis 2, such that STP is not positively linked with PCF.

The reason for the rejection of hypothesis 2 can be explained as SEM analyses the model as a whole, neglecting the importance of the individual relationships (Tomarken & Waller, 2005). One of the significant issues is that it tests the relationship based on the researcher’s assumptions and understanding. When presented with more complex material, the approach begins to struggle to comprehend. It is incapable of identifying additional contributing variables if the proposed ones do not contribute significantly.

RandomForest (RF) regression supervised learning technique

The following section discusses the RF approach, which divides the entire data set into training and testing datasets. In the first step, 50% of the data was trained, and the remaining 50% was used to test the resulting patterns. Consequently, the data were further regressed, and Table 4 provides a comprehensive data set with each item and factor used and labeled throughout the experiment. The table depicts eight final factors and items used to train the data set in preparation for the development of the final prediction model. Instead of linear regression, a random forest regressor was utilized since it makes no assumptions about the linearity of the connection between variables, collinearity, or homoscedasticity (Golino & Gomes, 2016). It is also the most effective supervised learning model algorithm (Hammou et al., 2019) for predicting unknown preferences. Four predictive models were created by training the data items on JS, OCBI, OCBO, and PCF, respectively.

The obtained results are accurately depicted (R2 value) in Table 5. With 85%, 96%, and 97% accuracy, JS, OCBI, and PCF may be predicted, respectively. The association between OCBO and data items can be eliminated because the data is inadequately trained to predict its nature. The model additionally incorporates the demographic factors of the participants and provides more explicit prediction for the determinants. Figure 3 depicts the bar plot of all the determinants, which provides a graphical depiction of the actual versus projected data (test data), demonstrating high levels of precision. Using the presented data set, our developed predictive models can anticipate the determinants. In addition, the predictive models identified specific items (e.g., T3, T4, T10) with uncertain preferences in anticipation of OCBO, JS, OCBI, and PCF.

Table 5.

Supervised learning model with R^2 value (accuracy)

Y	X (Predictors)	Root mean square error	R^2 value
JS	Age + TR1 + TR2 + TR3 + TR4 + OCBI	0.28725	0.85858
OCBI	Age + Gender + Industry Experience + T3 + T4 + T10 + T12 + T13 + TR1 + OCBO + JS	0.21128	0.9647
OCBO	Industry Experience + Earners + PCF + JS6 + JS8 + OCBI	1.02721	0.16702
PCF	Income group + T14 + JS6 + OCBO	0.1778	0.97503

Fig. 3.

Four figures show the bar plot between the actual data set and predicted data set between Employee ID and JS, OCBI, OCBO, PCF, respectively

This prediction technique provides a clearer picture than the conventional technique, which was unable to comprehend the complexity of the data and hence ignored the crucial functions of variables. Different demographic variables not considered by the traditional method played a significant impact in forecasting employee behavior, as demonstrated by the results. In addition, it added a layer of certainty to the already established dependencies between JS and OCB.

Previously, studies including STP, PCF, JS, and OCB examined a well-defined relationship; however, by machine learning modeling, we attempted to highlight the function of the determinants in predicting one another’s behavior. The degree to which these factors influence one another can aid in anticipating one another. Suppose, for instance, that the management is interested in determining whether or not the employee’s psychological contract has been met. In such situation, they can readily forecast it with the assistance of our model, and implement required measures.

Discussion

Volatility, Uncertainty, Complexity, and Ambiguity have become the major features of the current business environment (Bennett & Lemoine, 2014). Working in such a business climate requires organizations to prepare their staff for unforeseen challenges while also realizing how critical important an employee has become and how every organization’s operation necessitates a long-term and sustainable solution. One such solution is sustainability in every aspect of the organization (Mariappanadar, 2020; Aust et al., 2020; Stankevičienė & Savanevičienė, 2018). In order to equip their employees to be able to come up with sustainable solutions, their human resource strategies require a significant implementation surge. Furthermore, “training” is a component of HR policy that drives this implementation and prepares employees to confront any change (Merzbacher, 2022).

In the present study, we tried to evaluate the relationship of STP with JS, PCF, and OCB in the ongoing pandemic scenario. The study found a significant relationship between STP with JS and OCB but not with PCF. The investigation progressed and incorporated machine learning ideas to construct an all-encompassing prediction model for STP, JS, PCF, and OCB. The first hypothesis was supported revealing that employees were content even when working conditions changed and their exchange process altered but remained intact, underlining the role of SET (Cho & Yi, 2022). In such a volatile scenario, employees maintained their composure and went about their everyday duties. This demonstrates that the influence of STP on employees was significant enough that their satisfaction levels were unaffected by challenging circumstances. It helps to determine the extent to which sustainable practices affect employee dispositional traits. This could be due to the fact that different STP components focus on addressing different types of employee needs in training programs and providing different levels of training, and that the role of the organization in promoting training programs along with the working environment will influence employee satisfaction (Kwon, 2019). Another interpretation posits that the nature of STPs gives employees the impression that they are being cared for on an individual basis (Heimerl et al., 2020).

The second hypothesis showed that STP did not significantly impact employees’ psychological contracts as the exchange process had changed due to the pandemic. As organizations’ working environments transformed, previous promises became warped, resulting in a misunderstanding of the existing role of unwritten exchanges. Previous research highlights the impact of STP on PCF, but SEM fails to recognize it (Chinyamurindi, 2021; Jilani et al., 2020; Rao, 2021; Zhang et al., 2019a, 2019b). The one reason for insignificance might be the usage of the traditional technique; as stated earlier, it sometimes fails to acknowledge important relationships when the complexity of the data increases; whereas on the other hand RF technique shows that PCF is an important component in predicting OCBO in combination with other factors.

The third hypothesis was supported as STP also had a key influence in molding the behavior of employees within an organization, i.e., how they interact with their collaborators, mentors, co-workers, managers, and so on. First, advanced levels of training, such as STP, emphasized the professional and personal needs of the individual, fostering a sense of belonging towards the organization. Subsequently, they revert to exhibiting greater amounts of OCBO and OCBI (Lee, 2015). The second benefit of a highly aspirational and learning workplace is that it provides employees with a forum for communication, transparency, and social dialogue, which results in improved employment relationships and greater job satisfaction (Barrena-Martínez et al., 2019). Thirdly, training programs centered on talent, skills, performance reviews, and employee commitment were more strongly associated with OCBO and job satisfaction, as employees felt a strong sense of obligation towards their firm. These epiphanies demonstrated why employees cared about their organizations and coworkers during challenging times. Consequently, this highlights the STP’s strengths and the need to enhance them.

The second section identifies the relationship drivers through the predictive modeling technique. The study primarily highlighted individually which variables are helping in determining the satisfaction, psychological contracts, and behavior of the employees. It not only aims to recognize the variables affecting the organization’s current employee base, but it also goes a step further in preparing itself to understand the demands of future employees giving organizations an edge for future endeavors. The resulting models focused on revealing the hidden relationships between individual data items, such as working environment conditions, how an organization encourages employees to participate in various programs, or how strongly an employee believes that his work is beneficial to his health, work-life balance, etc. The predictive analysis presented a comprehensive model because it included demographic information. Since it incorporated demographic data, the predictive analysis produced a comprehensive model. In ascertaining an employees’ behavior, e.g., PCF was not able to establish a strong relationship via SEM whereas the predictive model unveils the factors that contribute in predicting this construct as well such as "income group," "T14", "JS6", and "OCBO". The model can be implemented in current organizational setups as the organizations can use it in plugins with their existing HRMS systems. In addition, the feedback from the training sessions can be supplied as raw data to the predictive analysis model to keep improving the accuracy.

Theoretical implications

Theoretically, this study demonstrates how STP helps redefine, comprehend, and modify the behaviors and attitudes of an organization’s personnel (Jeronimo et al., 2020). The findings highlight the increasing significance of STP and its influence on employee satisfaction, psychological contracts, and behavior (Yáñez-Araque et al., 2017). As a result of training, employees believe that their management or company cares about them and supports their careers, which considerably improves these behaviors. Thus, by preserving a work culture, i.e., fostering the growth of people, companies may respond to such challenging times more effectively (Pellegrini et al., 2018). The results contribute to our understanding of SET by illustrating the predictive power of the theory. It further offers the prospect of a future viewpoint of SET, which could be investigated further by researchers utilizing several variables of sustainable ways in order to develop a modern predictive model for HRMS.

Managers must place greater emphasis on tailored training programs, as they better prepare individuals psychologically for tough situations. When employees see that their managers have upheld their end of the agreement by giving specialized training, they engage in citizenship behaviors and improve their overall performance. Relatedly, SHRM attempts to build a win–win strategy by balancing the demands of both sides, i.e., the employees and the organization (Elkington, 1994).

Practical implications

As a manager, it is crucial to understand the needs of employees without compromising those of the organization. The generated predictive model will not only help comprehend the psychology of existing employees (Singh et al., 2022), but also the demands and expectations of future employees. It also gives managers an edge as they can devise a proactive strategy for employee productivity (Haller, 2022). When tested during these times, STP showed that employees felt satisfied and did not suffer any major breaches of their expectations from the organization. That helped people maintain their work-life balance such that the change of newer working policies did not enhance the panic (Akanji et al., 2022). Also, it will help the employees be better prepared for any such disturbances or problems. Covid-19 has made us realize that in this dynamic scenario, we need to be prepared for any circumstances, and we can help organizations overcome such troubles by training their employees.

Future research and limitations

Our findings, while novel and provocative, should be interpreted with caution, keeping in mind certain limitations of our study. First, our study was conducted during the Covid-19 era, and this should be borne in mind as the pandemic impacted the day-to-day operations of organizations well as employee interactions and performance (see, e.g., Zhang-Zhang & Varma, 2020). More specifically, our study investigated critical constructs such as job satisfaction, organizational citizenship behaviors and psychological contract fulfillment. Given that organizations had to resort to work-from-home and/or hybrid work almost overnight, it is clear that constructs such as JS, OCB, and PCF would have been impacted. As such, future scholars should examine our study relationships during the post-pandemic era and compare their findings with ours.

Second, due to the ongoing pandemic, our data could only be collected online, via email or through social media platforms, such as WhatsApp, Facebook, and LinkedIn. Future scholars should attempt to replicate our study using data obtained through alternate channels, such as face-to-face data collection by physically visiting the locations. In addition, scholars could concentrate on a particular industry or type of ownership to determine industry-specific behaviors with greater precision.

Third, as our data were collected via surveys sent to the participants, our findings could be impacted by self-selection bias. In future investigations, researchers could employ structured sampling and a multi-source, 360° data collection approach to control for this. Additionally, longitudinal studies can be undertaken to examine the degree of change in the behavior of individuals over time, thereby uncovering additional factors that can influence their behavior.

Finally, our study was conducted in India, an important global economy and a major player in Asia. However, the generalizability of the results to other countries in Asia and the rest of the world is questionable, given the unique context of the Indian economy and the Indian workplace. As such, we would like to urge future scholars to examine our model in other countries in Asia, and beyond. Such investigations would help increase confidence in our findings.

Biographies

Akriti Gupta

is currently a Senior Research Fellow and Ph.D. scholar in the Department of Management Studies at the Indian Institute of Information Technology, Allahabad. Her research interests include, but are not limited to, Organizational Behavior, Work Culture, Sustainable Human Resource Management practices, Strategic Human Resource Management practices, Job Satisfaction, and the VUCA model.

Aman Chadha

is currently a Senior Research Fellow and Ph.D. Scholar in the Department of Management Studies at Indian Institute of Information Technology, Allahabad. His research interests include Organizational Behavior, Human Resource Management practices, Sustainable Human Resource Management, Training And Development, Psychological Contract Fulfillment, Organizational Citizenship Behavior, and Emotional Intelligence.

Prof. Vijayshri Tiwari

is currently a Professor in the Department of Management Studies at the Indian Institute of Information Technology, Allahabad. She has completed her Masters’s in Psychology and HRM. Her career has made her increasingly specialise in managing complex academic deliverables and projects. Her research endeavours include her indulgence in the application of modern leadership methods, motivation, clarity in communication, Strategic HR, and personal inter-relation. Her background and interests in quality service delivery have also made her confident in handling students of different backgrounds and calibre and reconciling different approaches to otherwise common problems. Her research interests include Work Culture, leadership, Group Dynamics, Performance Management, Motivation, Strategic HR, Employer Branding, Sustainable Human Resource Management, Psychological Contract.

Arup Varma

(PhD, Rutgers University) is Distinguished University Research Professor and the Frank W Considine Chair in Applied Ethics at the Quinlan School of Business, Loyola University Chicago (USA). Dr. Varma’s research interests include performance appraisal, expatriate adjustment, and HRM issues in India. He has published over 100 articles in leading refereed journals such as the Academy of Management Journal, International Studies of Management & Organization, the Journal of Applied Psychology, Personnel Psychology, the Journal of International Management, the Journal of Business Research, and the Journal of World Business.

Prof Vijay Pereira,

PhD (United Kingdom) is Full Professor of International and Strategic Human Capital Management and Department Chair of People and Organizations department at NEOMA Business School, France and Distinguished Full Professor at Goa Institute of Management (GIM) and Indian Institute of Management, Kozikode. Professor Pereira is the 2nd ranked publishing scholar in business and management globally for the year’s 2021–22 and the highest ranked publishing scholar in Europe for the same period (P-rank). He was Associate Dean (Research) at the Australian University of Wollongong (Dubai campus). He holds adjunct positions of Full Professor at University of South Pacific, Fiji and at Universities of Portsmouth and Manchester. Professor Pereira is the Editor in Chief of the journal International Studies of Management and Organizations, the former Associate Editor (Strategic Management and Organizational Behavior) for the Journal of Business Research and the Global Real Impact Editor for the Journal of Knowledge Management. Prof Pereira is also on the editorial and advisory board for the journals Production and Operations Management and Journal of Management Studies (both listed in Financial Times ranking). He has a record of attracting funding and has published widely, in over 200 outlets, 20 special issues and 10 books, including in leading international journals such as the Human Resource Management, Harvard Business Review, MIT Sloan Management Review and Journal of Business Ethics (all Financial Times ranked). He has also published in the Academy of Management journals Academy of Management Perspectives and Academy of Management Discoveries. He was the elected Vice President of the Academy of International Business (AIB), Middle East and North Africa (MENA).

Declarations

Conflict of interest

The authors declare that they have no competing interests.