<![CDATA[Recognizing the varying learning styles of students is vital to creating customized educational approaches and maximizing academic success. While commonly used, conventional evaluation methods such as self-report surveys are frequently characterized by subjective biases and inconsistent accuracy. To address this limitation, this present study proposes an EEG-driven approach for learning style classification, specifically targeting the Active and Reflective dimensions of the Felder-Silverman Learning Style Model (FSLSM). Data was acquired from 14 participants using an 8-channel OpenBCI headset, with cognitive engagement stimulated through Raven’s Advanced Progressive Matrices (RAPM). Initially, the raw EEG data underwent bandpass filtering process purposely to remove noise. Subsequently, the data was divided into consecutive 1-second segments. For feature extraction, the CUSUM algorithm was employed, with an aim to effectively capture significant signal variations. These features were then fed into an LDA classifier for style discrimination. The performance evaluation revealed impressive results—98.26% accuracy in standard Train-Test validation, and an even higher 99.29% under LOOCV testing. Notably, our approach consistently outperformed existing techniques including 1-DCNN and TSMG across all metrics. Notably, computational efficiency and reliability were improved, with the "Odd-only" subset yielding peak accuracy (99.24%). These findings demonstrate that integrating EEG signals with conventional machine learning enables real-time, high-precision learning style detection. Additionally, this work addresses the computational constraints and dataset limitations observed in recent studies, providing a robust foundation for adaptive learning systems. It is recommended that future research explore larger, more diverse datasets and additional FSLSM dimensions to enhance generalizability and practical implementation of the research.]]>
Copyrights © 2025
