International Journal of Science, Engineering and TechnologyAuthors: Dr Tanu Gupta, Dr Mridula, Jaishree Goyal
Abstract: Small daily factors, such as attendance, study effort, family support, sleep, and co-curricular engagement, have a significant impact on student achievement. Often, these signals are only apparent after a student's grades drop, which makes timely intervention difficult. The purpose of this paper is to present a Student Performance Prediction System that combines an interpretable machine-learning pipeline with a web-based interface to provide prompt, student-friendly feedback. Two methods are used to model academic performance in dataset 1: (i) GPA prediction using regression and (ii) grade-category prediction using multi-class classification. We examine the performance of two powerful tree-based learners, Random Forest and Gradient Boosting, in a held-out test split. Gradient Boosting outperformed Random Forest (R2 0.924) in GPA regression with a R2 of 0.946 and RMSE of 0.212 GPA points. Gradient Boosting has the best macro F1 among tested baselines when it comes to grade categories, with the models achieving approximately 0.71 accuracy. Beyond model accuracy, the system focuses on usefulness: feature importance and correlation analysis are used to explain dominant drivers (e.g.Absences and study time are model signals that the web UI converts into short, actionable recommendations. We conclude by describing the complete implementation (React Vite frontend, FastAPI backend, REST endpoints) and discussing ethical considerations such as bias, privacy, and responsible communication of predictions.
DOI: http://doi.org/10.5281/zenodo.21455641
