Harnessing the Power of a Bidirectional Long Short-Term Memory-Based Prediction Model: A Case of Student Academic Performance

This paper aimed to develop a Bidirectional Long Short-Term Memory (BiLSTM) based model for predicting student academic performance using Principal Component Analysis (PCA) as a feature selection technique. The study was motivated by the concern about the high level and rates of failure of undergraduate students in Nasarawa State University Keffi, Nigeria, despite the University’s efforts to provide an enabling learning environment. Previous studies in this area focused on the program-based approach, which may not fit appropriately into the local educational system. Additionally, the measures of academic performance used in those reviews fluctuated, bringing about a lacking assessment of indicator factors in understudy information. As a result, a novel strategy was proposed in this study by combining improved methods for selecting features with a BiLSTM model that has been successfully used in a variety of contexts. Data collection, feature selection, and the use of a deep learning model to predict student grades were all part of the method used. The most relevant characteristics that can be used to predict student academic performance were identified using PCA methods. The utilization of this component determination method assisted with lessening the dimensionality of the information while protecting the main data, bringing about a more exact prescient model. The model exploratory outcomes showed that it yielded the best outcomes when compared with other studies in the same area, achieving an accuracy of 90.16%. The study’s findings could be useful for educators and administrators in Nigerian universities, contributing to improving the quality of education and reducing undergraduate student failure rates.