Fairness-Aware AI for Predicting Student Dropout and Enhancing Personalization and Collaboration in Learning Digital Learning

Abstract

Digital education faces the challenge of balancing personalized learning with equitable collaborative engagement while ensuring fairness in AI-driven systems. This study conducts a comparative analysis of machine learning models to predict student dropout likelihood in personalized learning environments, addressing the need for early identification of at-risk learners. Using the Kaggle "Personalized Learning & Adaptive Education" dataset, which contains 15 demographic, behavioral, and performance features, we evaluated eight machine learning models: Random Forest, XGBoost, CatBoost, LightGBM, SVC, Naive Bayes, KNN, and AdaBoost. The Random Forest model achieved the highest performance (Accuracy = 0.832, Precision = 0.843, Recall = 0.832, F1 = 0.831), demonstrating superior capability for identifying at-risk learners. Correlation analyses revealed that engagement metrics, particularly assignment completion and time spent on resources, were stronger predictors of academic success than demographic factors. A negative correlation was found between a reading/writing learning style and final exam scores, suggesting a need for more visual and interactive educational resources. These findings provide pedagogical insights and a foundation for designing effective intervention strategies in personalized education systems.