Feature Selection of Student Performance using Random Forest Algorithm

The challenge of enhancing predictive accuracy in student performance prediction takes center stage in this research endeavor. Exploiting the robust capabilities of the Random Forest algorithm and drawing from the renowned Open University Learning Analytics dataset, this study delves deep into the intricate realm of feature selection. Amidst the global academic landscape, the paper embarks on a comprehensive and rigorous journey, exploring an array of feature selection techniques. These include chi-square, information gain, correlation coefficient, variance threshold, Fisher score, RFE, RFECV, select from model, L1 regularization, decision tree importance, elastic net, SVM, PCA, select k-best, select percentile, L1 regularization CV, L2 regularization CV, and elastic net. A meticulous evaluation process unveils a definitive approach to feature selection, finely attuned to the nuances of international academia. Empirical findings cast the RFECV method as the unrivaled frontrunner, achieving an impressive accuracy rate of 94.75%. This methodological choice paves the way for selecting features that prominently impact predictive precision. The outcomes presented herein offer indispensable insights destined to empower educational practitioners on a global scale.