MACHINE LEARNING-BASED PREDICTION OF RURAL STUDENTS’ ACADEMIC PERFORMANCE IN URBAN UNIVERSITIES USING CHI-SQUARE FEATURE SELECTION AND BMGSVM-SMO

Abstract

Educational Data Mining (EDM) plays a critical role in identifying factors influencing student success and predicting academic outcomes. This research presents a comparative study on predicting student academic performance levels ('Low', 'Medium', 'High') using the SPD.csv dataset. The methodology involves robust data preprocessing (handling missing values, encoding, and normalization), feature selection using the Chi-Square test, and classification via three distinct models: Logistic Regression, Random Forest, and a novel hybrid approach, the Boosted Multi-Gradient Support Vector Machine optimized by the Spider Monkey Optimization (BMGSVM-SMO) algorithm. The results, evaluated based on Accuracy and Precision, indicate that the meta-heuristic-tuned BMGSVM-SMO significantly outperforms the baseline models, demonstrating the efficacy of integrating evolutionary computation for hyperparameter optimization in complex educational prediction tasks.