A multi-dimensional prediction system for students’ academic performance driven by deep learning

Abstract

The Academic Multi-Factor Prediction Net (AMP-Net) is a comprehensive deep learning framework that predicts student academic performance by integrating multiple data dimensions, unlike conventional prediction systems, which often overlook the interplay between engagement, emotional states, and socio-economic factors. AMP-Net considers historic grades, classroom engagement, socio-economic classification, and sentiment derived from students’ digital footprints. By fusing, normalizing, and transforming these diverse datasets into a unified vector, AMP-Net captures subtle non-linear relationships across heterogeneous data sources, enabling more accurate and holistic predictions. This framework allows for educational institutions to proactively identify at-risk students and design targeted interventions based on meaningful engagement indicators. The deep neural network architecture of AMP-Net ensures robust performance across multiple evaluation metrics. Experimental results demonstrate its effectiveness, with academic features achieving a reliability of 0.87 and an accuracy of 0.85, engagement data showing a scalability of 0.78 and a precision of 0.81, and socio-economic inputs contributing a recall of 0.82 and an AUC-ROC of 0.87. Sentiment analysis further enhances predictive power, with a low MAE of 0.13 and strong interpretability at 0.88. By leveraging these multidimensional insights, AMP-Net enables institutions to develop actionable, data-driven strategies that optimize resource allocation, support student success, and enhance overall academic outcomes, surpassing the predictive capabilities of previous models.