Enhancing Student Knowledge Assessment through CNN and Attention Bidirectional GRU using a Comparative Analysis

To enhance academic outcomes and reduce dropout rates, predicting student performance in educational environments is crucial. Curriculum creation, academic support intervention planning, and behaviour prediction are significantly improved by Student Knowledge Assessment. Technology-enhanced learning tools necessitate advanced predictive models because of the substantial data they generate. The proposed methodology commences with pre-processing procedures, including data cleansing, case folding, tokenisation, stop word elimination, negation conversion, and stemming. An innovative Adaptive Feature Selection (FS) Algorithm enhances robustness through the integration of five distinct algorithms, employing a normalised mean ranking and an ensemble-based initial feature ranking. They employ a CNN for dependable feature extraction and a BiGRU for the documentation of contextual details. Additionally, significant terms are emphasised by an attention method. The proposed model attains an accuracy of 96.76%, surpassing existing state-of-the-art techniques, as indicated by the experimental results. Its effectiveness in predicting academic performance and Student Knowledge Assessment is underscored by its elevated accuracy. This study improves academic interventions, customised learning, and curriculum development by integrating deep learning models with adaptive feature selection. The findings imply that the proposed model may serve as a valuable tool for educational institutions and educators in their efforts to enhance student performance.