Enhancing Student Retention: Predictive Machine Learning Models for Identifying and Preventing University Dropout

Student dropout in higher education poses a critical challenge with far-reaching academic and socio-economic consequences. This study proposes a machine learning-based multi-class classification framework to predict student status—categorized as dropout, enrolled, or graduate—by leveraging a comprehensive dataset that includes academic, demographic, and socio-economic variables. Feature selection techniques, including variance thresholding, SelectKBest, Principal Component Analysis (PCA), and Genetic Algorithms, were employed to enhance model efficiency and accuracy. An ensemble learning approach with multi-source data fusion yielded the highest performance, achieving an overall accuracy of 93.5%, thereby surpassing traditional predictive models. The methodology incorporates both classification and clustering techniques, with evaluation based on precision, recall, F1-score, and clustering indices to ensure robustness. Key predictors identified include academic performance, behavioral patterns, and socio-economic conditions, offering valuable insights for developing targeted intervention strategies. The proposed model demonstrates strong potential in mitigating dropout rates, with future work aimed at broader dataset generalization and enhanced model interpretability.