Artificial intelligence and machine learning offer promising solutions for predicting school dropout. This study proposes a novel approach by integrating psychosocial features, such as health and family relationships, into machine learning models for predicting secondary school dropout. Using an open-source dataset, eight classification algorithms were evaluated, with XGBoost achieving the best performance (accuracy: 96.64%, F1-score: 0.9636). The inclusion of psychosocial features improved predictive performance, notably increasing accuracy by 5.04% and recall by 3.71%. SHAP analysis revealed the importance of health and family support as key predictors. Compared to previous studies, this approach stands out by leveraging psychosocial data and explainable AI techniques.