STUDENT ACADEMIC PERFORMANCE ANALYSIS USING SUPPORT VECTOR REGRESSION AND MULTILAYER PERCEPTRON: AN EDUCATIONAL DATA MINING APPROACH

Abstract

Predicting student academic performance lays a foundation for data- informed educational decisions. This work uses the public Student Performance Data set from Kaggle website which contains 10,000 records and is used for predicting ‘Performance Index’. Median imputation, one hot encoding for categorical variables and feature standardization were used for preprocessing of data. The model was evaluated through 5-fold cross-validation, and the proportions of training and testing data were set at 80:20 in each fold. Two different regressi on models were utilized: Support Vector Regression (SVR) with RBF kernel and a Multilayer Perceptron(MLP) consisting of two hidden layers(128–64 neurons). Both models achieved excellent prediction accuracy. SVR achieved an MAE of 1.6653, RMSE of 2.0991 and R²=0.9881 whereas MLP slightly performed better than SVR with a MAE 1.6596, RMSE of 2.0872 and R²=0.9882. Learning curve analysis showed stable convergence with little overfitting. The results show their efficiency and they are both kernel-based and neural network-based methods to predict academic performance. Future work will need to test the models on much more diverse data sets and may incorporate further context variables to improve model robustness and interpretability.