Enhancing student performance prediction using an optimized RNN-LSTM model: a step towards personalized education

Learning Analytics is an emerging field that strives to analyze and predict trends in student performances over time. With the increased usage of Learning Management Systems (LMS), which are equipped with learning content, interaction, and assessment aligned with educational objectives, advanced machine learning techniques such as Long Short-Term Memory (LSTM) are being used to analyze sequential data. This research aims to assess the efficiency of LSTM in student performance prediction by comparing its accuracy with other baseline models such as Convolutional Neural Networks (CNNs) and Gated Recurrent Units (GRUs). The proposed optimized recurrent LSTM model outperforms CNN and GRU baseline models in prediction accuracy and efficiently captures temporal dependencies in sequential student academic data. To enhance training stability and predictive performance, the model was trained using Binary Cross-Entropy loss and validation-based parameter selection. The research indicates that such a model has the ability to be used in personalized learning by adjusting learning paths dynamically to improve student performances.Experiments were conducted using a synthetic dataset designed to simulate realistic academic progress patterns, demonstrating the model’s predictive capability.