Stress is a growing mental health issue in the world, as it is predisposed to the lifestyle factors that are amenable to modification. This paper applies machine learning to identify the main predictors of stress and assess model performance. A sample of 2,000 participants who had 11 lifestyle characteristics, such as sleep, digital habits, physical activity, workload, and environment, was examined. Median (1.40) was used to binarize stress levels. The 5-fold stratified cross-validation was used with hyperparameter optimization to train four models: Logistic Regression, Random Forest, Support Vector Machine (SVM), and XGBoost. The performance of all models was high (ROC-AUC > 0.97), and the highest accuracy was demonstrated by Logistic Regression (91.75), and ROC-AUC (0.983). Analysis of feature importance showed that daily pending tasks had a contribution of approximately 34.6 percent to predictions, which was far more than other factors. The moderate effect (7-8% of influence) was on sleep variables, and the remaining features had a small effect (less than 0.5%). These results confirm the Task Load Hypothesis, which underlines workload management as the most effective method of reducing stress
Keywords: Stress prediction, feature importance, lifestyle factors, machine learning, mental health analytics, workload management
