International Journal of Science, Engineering and TechnologyAuthors: P.Latha Gowri
Abstract: Predicting student academic performance is a valuable application of machine learning that helps educational institutions identify at-risk students and provide timely interventions. This paper presents a study on using Linear Regression and Multiple Linear Regression models to predict student final examination marks based on factors such as study hours, attendance percentage, previous semester marks, and assignment scores. A dataset of 200 student records was used to train and evaluate the models using Python's scikit-learn library. Simple Linear Regression was first applied using study hours as the sole predictor, followed by Multiple Linear Regression incorporating all four predictor variables. Model performance was evaluated using R-squared, Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). Results show that the Multiple Regression model (R² = 0.847) significantly outperforms the Simple Linear Regression model (R² = 0.612), indicating that combining multiple academic and behavioral factors provides a more accurate prediction of student performance. Previous semester marks and attendance percentage were found to be the strongest predictors. The study concludes that multiple regression-based predictive models can serve as an effective early-warning tool for academic institutions.
DOI: http://doi.org/10.5281/zenodo.22708067
