AUTOMATED LEARNING STYLE CLASSIFICATION USING AUTOENCODERS AND FELDER-SILVERMAN MODEL FROM STUDENTS' DIGITAL FOOTPRINTS

Abstract

This study investigated machine learning approaches that utilize students' digital footprints to classify learning styles and compared them with the established Felder-Silverman model. A dataset of 120 students with detailed digital footprints was collected from an online course-management system. The students completed the Felder-Silverman learning style questionnaire. In addition to employing principal component analysis (PCA) for feature extraction, an autoencoder neural network was proposed in this study for unsupervised learning of an informative latent feature representation from the raw footprint data. The encoder features from the autoencoder and the principal components from PCA were input to the K-means and hierarchical clustering algorithms to classify students into learning styles. The predicted clusters were compared with true Felder-Silverman labels by calculating accuracy, precision, and recall. The results showed that the K-means algorithm achieved a significantly higher accuracy of 89.7% with autoencoder features compared with 85.2% with PCA features. The autoencoder model outperformed both PCA and traditional classifiers, such as decision trees. These results demonstrate the potential of using autoencoders to enhance feature learning from digital footprints for more precise learning-style classification.