Utilizing Support Vector Machine and Dimensionality Reduction to Identify Student Learning Styles within the Felder-Silverman Model

Abstract

This research explores the impact of questionnaire structure on the accuracy of learning style classification, focusing on the optimization of the Felder-Silverman Learning Style Model (FSLSM) using advanced machine learning techniques. By employing Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE) for dimensionality reduction, the study identifies and retains the most informative variables from the original 44-question FSLSM instrument. These refined features are then processed through a Support Vector Machine (SVM) algorithm to evaluate classification performance across various core-to-secondary item ratios. Results indicate that the most optimal configuration—produced through the combined PCA-t-SNE reduction—achieved a peak accuracy of 89.54%, surpassing other configurations and highlighting the effectiveness of selective question modeling. This approach not only enhances prediction accuracy but also introduces a more efficient and streamlined FSLSM formula, reducing redundancy without compromising diagnostic precision. The study contributes to educational data mining by presenting a data-driven strategy for learning style assessment and offers practical implications for the development of adaptive, personalized learning systems grounded in statistically validated models.