Predict Students’ Performance and Optimize Preparatory Leave: An Integrated Deep Learning and Dynamic Programming Approach

Data-driven decision-making is critical for accurate outcomes across various fields, yet its application in the education sector remains limited. With the increasing capabilities of machine learning models, the true value lies in enhancing prediction accuracy through their optimized application. This paper employs prescriptive analytics to construct an optimal preparatory leave schedule within a two-stage framework—first predicting and then optimizing. Key features influencing academic grades are identified through a comprehensive literature review, followed by an analysis of their correlations with academic performance. Subsequently, various Machine Learning, Deep Learning and Graph Neural Networks algorithms are implemented to predict student grades. Among them, artificial neural networks (ANN) emerge as the most effective, exhibiting the highest accuracy (0.730) and the lowest mean absolute error (0.096). The predictions from ANN serve as inputs for creating an optimal schedule. Three distinct algorithms are used for optimal scheduling, brute force programming, dynamic programming, and branch-and-cut. Dynamic Programming outperforms the other approaches for optimal schedule-making regarding time, memory, and computational resources. In general, utilizing ANN for grade prediction, coupled with Dynamic Programming for effective scheduling of preparatory leave, ensures optimal results.