<![CDATA[The rapid growth of online learning environments has increased the importance of Learning Management Systems (LMS) as a rich source of behavioral data for learning analytics. One learner characteristic that strongly influences learning effectiveness is learning style; however, traditional questionnaire based identification approaches suffer from subjectivity, limited scalability, and static representation. To address these limitations, this study proposes a machine learning-based approach for automatic learning style classification using LMS behavioral data grounded in the Felder–Silverman Learning Style Model (FSLSM). This study utilizes LMS activity log data collected from Universitas Siber Asia over three academic years (2022–2024). The dataset consists of 5,633 student interaction records with 72 raw behavioral attributes, which were preprocessed, aggregated, and transformed into 12 representative behavioral features reflecting students’ interactions with learning materials, assessments, discussions, multimedia resources, and navigation patterns. A rule-based FSLSM mapping mechanism was applied to generate 16 learning style profiles, which were treated as targets in a single-label classification setting. Support Vector Machine (SVM) and Gradient Boosting (GB) classifiers were implemented and optimized using feature selection and GridSearch-based hyperparameter tuning. The dataset was divided into 75% training data and 25% testing data using a stratified split to preserve class distribution. Experimental results show that Gradient Boosting consistently outperforms SVM across all evaluation metrics. The GB model achieved an accuracy of 0.84 and a macro F1-score of 0.79, demonstrating strong generalization capability and robustness to class imbalance. In contrast, SVM exhibited lower and less stable performance, particularly on minority learning style classes. These findings confirm that ensemble-based methods such as Gradient Boosting are more effective for LMS-based single-label learning style classification and support the feasibility of automatic FSLSM-based learning style detection for data-driven adaptive learning systems.]]>
