This work investigates using machine learning models and feature selection techniques, including SHAP (SHapley Additive exPlanations), to predict student stress levels. We applied five classifiers to a dataset of student stress factors, encompassing physiological, psychological, academic, environmental, and social variables. Our analysis addressed three main research questions: (i) the effectiveness of ML models in predicting stress, (ii) the consistency of feature importance across classifiers, (iii) the viability of SHAP as a feature selection method, and (iv) the alignment of stress factors identified by machine learning techniques with established psychological and biopsychosocial frameworks. Results indicate that physiological and psychological factors, such as blood pressure, sleep quality, anxiety level, and self-esteem, consistently influence models. At the same time, academic performance also contributes significantly to stress prediction. SHAP performed comparably to traditional feature selection methods, enhancing model interpretability by identifying the most impactful features and improving classification performance. All classifiers achieved high F1 scores, demonstrating robustness in stress prediction. Results underscore the importance of integrating data-driven approaches with domain expertise to analyze stress factors comprehensively.