Optimization of medical check-up patient segmentation for personalized healthcare services

Abstract

Medical check-up (MCU) services commonly apply uniform examination pathways that do not reflect individual variations in demographic and lifestyle risk factors. This study aims to develop an unsupervised learning model that enables early personalization of MCU recommendations using non-laboratory pre-screening data. A dataset of 11,061 participants was processed following the cross-industry standard process for data mining (CRISP-DM) framework, with median imputation, outlier removal, and min–max scaling applied to enhance data quality. K-means and agglomerative hierarchical clustering (AHC) were evaluated using Silhouette score and Davies–Bouldin index, and cluster characteristics were examined based on age, body mass index (BMI), sex, smoking, alcohol consumption, exercise, and family medical history. K-means with four clusters provided the most stable segmentation (Silhouette 0.366; Davies–Bouldin index 1.21), producing clinically interpretable profiles ranging from high-risk groups marked by smoking and elevated BMI to lower-risk individuals with active lifestyles. These findings demonstrate that simple pre-screening variables can support meaningful early risk stratification and guide personalized MCU pathways. The study also outlines an integration approach for embedding the clustering model into electronic medical record (EMR) systems to enable real-time deployment within clinical workflows.