Optimizing K-Means Clustering through Distance Metric Simulation for Strategic Enrollment Segmentation in Private Universities

Abstract

K-Means clustering is a widely used unsupervised learning technique for identifying patterns and grouping data based on feature similarities. However, the effectiveness of K-Means significantly depends on the choice of distance metric. This study conducts a comprehensive simulation to evaluate and compare the performance of four distance metrics—Euclidean, Cityblock (Manhattan), Canberra, and Mahalanobis—in the context of strategic market segmentation for private universities. The dataset includes simulated and institutional data incorporating variables such as account creation, registration, graduation, student performance (social, science, and scholastic scores), income, and geographic distance. The results indicate that Euclidean and Cityblock distances yield efficient and interpretable clusters with low computational costs, whereas Mahalanobis distance, despite its capacity to model covariance, introduces computational overhead without proportional improvement in segmentation quality. Interestingly, Canberra distance produces compact clusters but offers no significant gain in separability. From the resulting segmentation, two clusters emerge as high-potential targets for marketing strategies: Cluster 0 (high-income and distant students) and Cluster 1 (diverse academic and socioeconomic profiles). The findings highlight the importance of aligning distance metric selection with specific clustering objectives and offer practical insights for data-driven strategic enrollment planning in private higher education institutions.