Analyzing diabetes patient data through k-means and principal component analysis

Communication and information technology advancements have significantly impacted various aspects of life, including healthcare. The proliferation of digital health data has enhanced the decision-making and management of chronic diseases such as diabetes. This study aims to analyze the characteristics of diabetes patients using k-means clustering combined with Principal Component Analysis (PCA) for dimensionality reduction. Clustering is a method within data mining that arranges data into clusters, aiming for similarities among members within each cluster. The process involves preprocessing, including data transformation, outlier removal, and min-max normalization. Clustering was verified with and without PCA, which is optimized by selecting the best k value based on silhouette scores. The result of the experiment showed that the highest silhouette score was obtained at k=2. The number of PC in PCA was determined by calculating variance values and examining the cumulative variance results. The best silhouette score of 0.67 was achieved with a 70% threshold and two PC at k=2. This score significantly improved compared to clustering without PCA, where the silhouette score was 0.56. This analysis revealed two distinct patient groups characterized by body mass index: one with ideal weight and the other with overweight. These findings highlight the potential for developing more precise and targeted care planning programs.