AI Application: Clustering E-Learner Profiles for Personalized Education: A Comparative Study Using K-Prototypes, Hierarchical, and HDBSCAN Algorithms

The research conducts an extensive clustering examination of student profiles through analysis of $\mathbf{1 4, 1 0 1}$ learner records that contain information across 21 demographic and behavioral engagement variables. The study aims to create meaning-based student groups to uncover learning disengagement patterns and tailor educational support measures. A comparison was conducted between K Prototypes, Hierarchical Clustering, and HDBSCAN to achieve this goal. Prototypes clustering generated three data segments of students, containing $3,521,5,115$, and 3,364 members, thus providing balanced and understandable cluster structures. The Ward linkage in the hierarchical clustering method produced three clusters with sizes $5,983,3,698$, and 2,319, establishing intense yet unstable groupings. HDBSCAN identified a substantial cluster of 11,507 students, alongside two meaningful clusters of 452 and 41 students, due to its unsupervised and density-based approach. K-Prototypes emerges as the best solution for clustering educational data containing both categorical and numerical features because it effectively processes mixed-type data before clustering, without requiring strong preprocessing steps. The created clusters serve education policymakers, academic advisors, and elearning platforms to develop retention-focused plans and individually customized educational approaches. The research developed an AI solution for an easily replicable workflow that supports learner segmentation, enabling data-driven academic learning and addressing the behavioral gap in analytic systems dedicated to higher education.