A Web-based Decision Support Platform for Student Performance Prediction using Machine Learning

Abstract

Background of study: Students confront more complicated academic decisions, ranging from course selection to study planning, but lack individualized, data-driven support. While machine learning has shown potential in forecasting performance, most educational technologies are institution-specific, technically obscure, or isolated from real-time student demands.Aims: This project develops, assesses, and implements a web-based Decision Support Platform that uses machine learning to provide students with individualized, real-time academic success forecasts and practical advice.Methods: The study tested seven regression models using RMSE, MAE, and R² on a 20% holdout set.  Nineteen behavioral, socioeconomic, and academic characteristics were preprocessed, and the most important predictors were statistically rated.  The dataset included 10,000 high school pupils.  The best-performing model was incorporated into a dynamic React-Flask web interface to enable real-time prediction.Result: Among the tested models, LightGBM outperformed all other options with the best prediction accuracy (R2 = 0.730, RMSE = 1.954).  Prior scores, study hours, and attendance were important predictors. With a sub-second latency, the deployed platform was able to produce real-time predictions based on user input.Conclusion: In conclusion, our findings indicate that academic planning may become insight-driven rather than intuition-based with the use of LightGBM-powered decision assistance. This initiative bridges the gap between educational machine learning research and equitable, real-world effect by putting predictive analytics in the hands of students, enabling them to make proactive, well-informed decisions about their academic futures.