Combining Random Forest, Neural Networks, And Association Rules For Student Grade Prediction And Course Recommendations

Abstract

This study proposes an ensemble-based approach for predicting university student performance and recommending optimal course combinations. The approach integrates Random Forest (RF), Long Short-Term Memory networks (LSTM) with Attention mechanisms, and Association Rule Mining (ARM) to address both individual academic forecasting and institutional course planning. RF is employed for effective feature selection and classification, while LSTM-A captures temporal patterns in students' academic trajectories. ARM is used to extract interpretable associations between course groupings and performance trends. The dataset contains detailed transcript and study plan records from 107 undergraduate students across 17 semesters. The experimental evaluation shows that the ensemble model achieves an accuracy of 82% and a macro-F1 score of 80%, outperforming traditional machine learning techniques. Additionally, the framework successfully identifies at-risk students with 85% accuracy in early semesters, supporting its potential use for academic advising and early intervention strategies.