Machine Learning-Based Prediction of Student Academic Performance Using Comparative Regression Models

Articles in Press

Document Type : Research Paper

Authors

1 Department of Computer Engineering, College of Engineering, University of Diyala, Diyala, Iraq.

2 Electronic Computer Centre, University of Diyala, Diyala, Iraq.

3 Department of Networks and Computer Software Techniques, Amarah Technical Institute, Southern Technical University, Basra, Iraq.

4 Engineering School of Digital Technologies, Yugra State University, Khanty-Mansiysk, Russia.

Abstract

Student academic performance has become a key area of concern in educational data mining and the accurate prediction of student performance has enabled timely interventions and improved learning results. This paper presents an in-depth analysis of ElasticNet, Ridge Regression, Gradient Boosting, Random Forests, K-Nearest Neighbors (KNN), and Decision Tree machine learning algorithms to predict student academic performance based on empirical data. Our experimental design will focus on early academic and behavioral outcomes, including assignment grades during the past four months, midterm grades, and monthly hours of attendance in the classroom and study sessions, and will be aligned to the scientific design principles that reduce the risk of future outcome leakage and related variables. The models are assessed based on usual measures of regression tasks, such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination (R2), and cross-validation to test the ability to generalize. The findings reveal that regularized regression models, namely ElasticNet and Ridge, are more predictive than others, with the R2 values approaching 0.5 and cross-validation stability being consistently high. Gradient Boosting and Random Forest show similar performance, but KNN and Decision Tree have rather low accuracy. Further, exploratory data analysis indicates that academic achievement measures provide a more accurate forecast compared to the use of behavioral characteristics. These results highlight the importance of carefully selecting variables and modeling approaches to ensure accurate forecasts of student performance. The findings are of great importance to both educational practitioners and policymakers as they give guidance on how institutional data can be used to improve evidence-based approaches to improve student outcomes.

Keywords

Main Subjects

Computational Methods for Differential Equations

Articles in Press, Accepted Manuscript
Available Online from 17 August 2026

Files

History

  • Receive Date: 04 July 2026
  • Revise Date: 31 July 2026
  • Accept Date: 16 August 2026

Share

How to cite

Statistics