Abstract
Forecasting student performance with precision in the educational space is paramount for creating tailor-made interventions capable to boost learning effectiveness. It means most of the traditional student performance prediction models have difficulty in dealing with multi-dimensional academic data, can cause sub-optimal classification and generate a simple generalized insight. To address these challenges of the existing system, in this research we propose a new model Multi-dimensional Student Performance Prediction Model (MSPP) that is inspired by advanced data preprocessing and feature engineering techniques using deep learning. We developed a method that targets the common issues associated with educational datasets over imbalanced and temporal settings which is also explainable through AI features. Moreover, through adaptive hyper-parameter tuning and advanced graph neural network layers in the MSPP model allow to make output more dense representation for predictions resulting more accurate classification. The experiments results show that MSPP outperforms the other EAI&ML, MTSDA, XAI, DGNN and DLM with high accuracy 76 %, precision score of 0.79 and macro F1-score of 0.73. The model also helps to bring down the False Positive Rate (FPR) substantially at a 0.15 level, which ensures more reliable predictions for student classification.•The model of the MSPP includes contextual information and multi-layered analysis in order to improve prediction accuracy, placing a sound basis for predicting students in different performance categories such as distinction, pass, fail or withdrawn.•Our approach is obviously to generalize and extract those sparse, heterogeneous academic data in the form of structured training record using domain specific preprocessing integrating with multi-class classification mechanisms that improves on precision-recall across multiple categories.