Predicting Student Performance Using Deep Learning and Machine Learning Techniques
Mohammad AlShaikh-Hasan, Gheorghiță Ghinea
Abstract
Accurately predicting student performance in higher education is crucial for early intervention and personalized support. In this paper, we propose a comprehensive framework that combines both traditional machine learning (ML) algorithms and deep learning (DL) architectures (LSTM and CNN) to forecast student academic outcomes. Our approach integrates demographic and academic registration data (REG) with course-level Intended Learning Outcomes (ILOs). Experimental results on a real-world dataset of 952 students demonstrate that models trained on the combined REG+ILOs features outperform those using only registration data. Among ten evaluated models—eight ML classifiers and two DL models—XGBoost achieved the highest accuracy of 82%, surpassing other ML methods such as Random Forest, while LSTM showed competitive recall and AUC for deep learning approaches. These findings underscore the value of incorporating ILO data into predictive pipelines and the robust performance of gradient boosting models for moderate-sized structured datasets.