Explainable machine learning for student dropout prediction and tailored interventions in online personalized education

The prediction of student dropout and development of specific intervention strategies continue to be major obstacles for contemporary educational systems operating through online personalized learning platforms. This study develops explainable AI (XAI) in education framework which predicts student dropout and generates specific interventions for online personalized educational settings. This research investigates three vital elements which include (i) studying how predictive learning analytics support AI model stability and generalization performance across various educational fields and (ii) adding student learning preferences as additional contextual data and (iii) building personalized explanations for students to help teachers develop intervention strategies. We trained and tested two ensemble algorithms (Random Forest and XGBoost) with a publicly available dataset which included demographic information and learning patterns and individual preference data. To improve model transparency, we applied SHAP for student retention analysis to provide both global and instance-level interpretations of model decisions, enabling educators to understand the drivers of predicted dropout risk. The models demonstrated strong performance across courses, with the highest accuracy observed in Data Science scenarios and lower performance in Web Development. The model applied learning-style characteristics to make decisions, yet these characteristics did not produce any significant variations in student dropout rates. The research findings show that learning-style characteristics failed to produce any significant predictive results, which confirms earlier studies that questioned their usefulness in academic environments. Notably, the predictive models identify relationships between variables, but they do not show how these variables affect each other, so users need to understand these results as machine-learning-based decision support rather than following causal instructions. The dataset comes from one public database, which restricts the ability to generalize the results and needs verification through multiple institutional datasets. Future systems require specific monitoring systems for fairness protection during the deployment of demographic variables. The research data show that explainable AI systems help predict student dropouts and create data-based intervention plans, which help build equitable online learning spaces.