Predicting Student Performance Using Deep Learning Models: A Comparative Study of MLP, CNN, BiLSTM, and LSTM with Attention

Abstrak

This study aims to predict student performance using deep learning models, including Multilayer Perceptron (MLP), Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Long Short-Term Memory with Attention (LSTM with Attention). The dataset comprises student demographic and educational factors, and the models are evaluated using metrics such as MAE, RMSE, R², MSLE, and MAPE. The results show that the CNN model outperforms other models, achieving the highest accuracy in predicting student test scores. The MLP model also performs well, while the BiLSTM and LSTM with Attention models exhibit lower predictive performance. High MAPE values across models suggest a need for alternative metrics in future research. This study highlights the importance of selecting suitable model architectures for predictive tasks in education, emphasizing the effectiveness of convolutional layers in capturing complex patterns.