Deep Learning-Based Student Performance Prediction And Analysis

Authors: Lavanya CM, M. Karthiyayini

Abstract: Predicting student performance accurately is vital in early detection of high-risk students and intervention in academic activities in higher education. This paper provides a systematic review on deep learning techniques used in student performance prediction. Recent developments in deep learning techniques such as hybrid stacked ensemble model, Gated Long Short-Term Memory network, Bidirectional LSTM with SHAP interpretability and integrated feature transformer networks have been discussed. The results of the study indicate that ensemble models utilizing several deep learning algorithms with performance-based weight give remarkable performance, with recall rate of 98.26%, precision rate of 99.51%, and F1-score of 98.88% in identifying at-risk students. Deep Learning based Gated LSTM models which use Dove optimization method gives 98.85% classification accuracy. Hybrid Deep Learning models, which integrate time-based behavioral patterns with static attributes of students give impressive results with various educational datasets. It has been shown that Deep Learning models provide better results compared to machine learning algorithms with accuracy rates over 97%. Data imbalance, lack of interpretability and generalization of educational data pose challenges.

DOI: http://doi.org/10.5281/zenodo.21338215