Support Vector Machine for Predicting Student Dropout Under Different Normalization Methods

Abstract

Student dropout in universities brings significant challenges that impacts both individual futures and institutional effectiveness. Early prediction of potential dropouts is crucial for timely intervention, but it is complex because of the nature of the problem influenced by diverse socioeconomic factors. This paper utlizies Support Vector Machines (SVMs) to predict student dropout with an emphasis on exploring the efficacy of various data normalization methods to optimize prediction accuracy. Using a dataset from the UC Irvine repository, this study compares 9 different normalization techniques such as Min Max Scaler, Standard Scaler, and Power Transformer, among others, to determine their impact on the predictive performance of SVMs. Results demonstrate substantial variations in model accuracy depending on the normalization method used to show the importance of detailed selection of data preprocessing techniques. The best normalization method was the One Hot Scaler which produced an average F1 score of 0.779. This work enhances the ability to identify at-risk students earlier but also the understanding of how data normalization influences predictive modeling in educational settings.