Prediction of Employee Absenteeism Based on Perceived Stress, Work–Family Conflict, Mental Health, and Procedural Justice Using Supervised Learning Algorithms

Abstract

The present study aimed to develop and compare predictive models of employee absenteeism based on perceived stress, work–family conflict, mental health, and procedural justice using supervised learning algorithms. This quantitative cross-sectional study was conducted among 312 employees working in public and private organizations in Tehran, selected through stratified random sampling. Data were collected using standardized instruments including the Perceived Stress Scale, Work–Family Conflict Scale, General Health Questionnaire (GHQ-28), and Procedural Justice Scale, all of which have established validity and reliability in previous studies. Employee absenteeism was measured based on self-reported absence frequency. Data analysis was performed using SPSS-27 and Python-based machine learning frameworks. Initially, descriptive statistics and correlation analyses were conducted, followed by the implementation of supervised learning algorithms including multiple linear regression, Random Forest, Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost). The dataset was divided into training and testing subsets (80/20 split), and model performance was evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and coefficient of determination (R²). Hyperparameter tuning and cross-validation techniques were applied to optimize model performance. The results indicated that perceived stress, work–family conflict, and poor mental health were positively and significantly associated with employee absenteeism, while procedural justice showed a significant negative relationship. Among predictors, mental health demonstrated the strongest effect, followed by work–family conflict and perceived stress. The regression model explained 42% of the variance in absenteeism (p < 0.001). Among machine learning models, XGBoost exhibited the highest predictive performance (R² = 0.63), outperforming Random Forest (R² = 0.58), SVM (R² = 0.55), and linear regression (R² = 0.42). Feature importance analysis revealed that mental health was the most influential predictor, followed by work–family conflict, perceived stress, and procedural justice. The findings highlight the critical role of psychological and organizational factors in predicting employee absenteeism and demonstrate the superiority of supervised learning algorithms, particularly XGBoost, in modeling complex relationships among variables. The study underscores the importance of integrating mental health support, work–family balance strategies, and fair organizational practices to reduce absenteeism and improve workforce productivity.