Predictive Modeling of Higher Education Dropout Using Deep Neural Network

Student dropout rates have become a critical concern in contemporary higher education, affecting not only the personal and professional futures of students but also undermining the reputation and operational efficiency of educational institutions. Students who discontinue their studies often face significant economic disadvantages and reduced career prospects, while institutions grapple with the broader consequences of decreased retention. Addressing this issue requires effective predictive solutions to enable early intervention. In response, this study proposes a novel approach centered on a Hybrid deep learning model, developed alongside RNN and LSTM models for comparative analysis. The Hybrid model stands out by classifying input features into numerical and categorical groups, processing them separately-numerical features through a Deep Neural Network (DNN) and categorical features through embedding layers combined with DNNbefore integrating their outputs for final prediction. Experimental results clearly demonstrate the superiority of the Hybrid model, achieving an accuracy of 90 %, an F1-score of 90.43 %, and a ROC-AUC of 94.36 %, significantly outperforming the baseline models. These findings highlight the Hybrid model's exceptional capability in predicting academic success and identifying at-risk students, offering not only accurate predictions but also deeper insights into the underlying factors affecting student outcomes. The proposed approach provides a strong foundation for developing targeted strategies to enhance student retention, and future work could explore integrating real-time academic data and deploying more sophisticated architectures to further refine prediction performance.