<![CDATA[This study is motivated by the need to group policyholders based on health risk levels and medical records to inform the formulation of health insurance premium policies. The methodology employs the K-Medoids algorithm combined with Gower Distance to process a mix of numerical and categorical variables, while the Silhouette Score is used to determine the optimal number of clusters. The dataset comprises 7,500 policyholders, with variables representing health risks and medical history specifically BMI, smoking status, physical activity level, alcohol consumption, chronic diseases, annual doctor visits, and a history of hospitalization in the previous year. The study yielded a Silhouette Score of 0.093783 (for k=3), indicating the data is best divided into three clusters. The analysis identified three distinct health risk and medical record groups: Cluster 2 (low risk), containing 3,431 policyholders with the lowest medical costs, chronic disease counts, and healthcare service utilization; Cluster 0 (medium risk), containing 1,975 policyholders characterized by the presence of chronic diseases leading to increased healthcare utilization despite having a normal BMI; and Cluster 1 (high risk), containing 2,094 policyholders with obese BMIs, the highest number of chronic diseases, and the highest medical costs, thereby presenting a higher potential risk for claims. The findings demonstrate that clustering using K-Medoids with Gower Distance can support the development of more proportional, risk-based health insurance premium policies.]]>
Copyrights © 2026
