Ai-Based Student Performance Prediction: A Systematic Review And Meta-Analysis

Abstract

Student performance prediction has emerged as a critical challenge in contemporary educational systems, 
necessitating the development of intelligent, data-driven solutions to identify at-risk learners and optimize 
academic outcomes. This paper presents a comprehensive review and meta-analysis of artificial intelligence (AI)
based approaches applied to student performance prediction, synthesizing findings from seminal and recent 
studies published between 2010 and 2024. The review systematically examines machine learning algorithms 
including decision trees, random forests, support vector machines, neural networks, and deep learning 
architectures deployed across diverse educational datasets encompassing demographic, behavioral, academic, 
and socio-economic variables. Through rigorous analysis of prediction accuracy, model interpretability, feature 
importance, and deployment contexts, this paper identifies prevailing methodological paradigms, critical 
research gaps, and emergent directions in the field. The meta-analysis reveals that ensemble methods and deep 
learning models consistently achieve superior predictive performance, with accuracy rates ranging from 75% to 
98% across reviewed studies. Furthermore, the study highlights the growing significance of learning management 
system (LMS) data, student engagement metrics, and multimodal data fusion in enhancing model robustness. 
Ethical considerations, including algorithmic bias, data privacy, and model transparency, are also critically 
examined. The findings underscore the transformative potential of AI-driven educational analytics while 
advocating for standardized benchmarking and inclusive dataset curation to ensure equitable and generalizable 
prediction systems.