Behavioral Clustering for Adaptive Learning: A Data-Driven Alternative to Static Learning Style Models

The existing predetermined paradigms of learning styles, including the Felder-Silverman Learning Style Model (FSLSM), VARK, Kolb’s experiential learning theory, and the Honey and Mumford model, have found significant application in personalized e-learning settings. However, these models typically rely on fixed, self-reported surveys that are not validated against actual learner behavior. This research addresses this shortcoming by conducting a behavioral analysis based on engagement data within a Learning Management System (LMS), incorporating elements such as content interaction, forum participation, and assessment performance. The K-Means++ clustering algorithm was employed to cluster learners and uncover latent behavioral profiles, which were then empirically compared with conventional models of learning styles to evaluate alignment. The FSLSM exhibited the strongest level of correlation with the behaviorally derived clusters (ARI = 0.87; NMI = 0.81), suggesting that it might encapsulate some persistent behavioral tendencies. But some key differences emerged in terms of time-on-task dynamics, student interaction behavior, and patterns of stress, none of which are wrapped within the FSLSM framework. This suggests that behavioral clustering describes actionable insights beyond profiles, which are static and self-reported, and allow for adaptive interventions responding to the real-time state of the learner.