Gaussian Mixture Model as an Effective Approach for Soft Clustering and Outlier Detection in Health Data

Identifying groups of individuals with similar health characteristics is essential for understanding population risk patterns and detecting vulnerable groups. However, health data often contain unusual records caused by measurement errors, input error, or uncommon health conditions. To address these challenges, this study applies the Gaussian Mixture Model (GMM) as a probabilistic clustering approach to analyze employee health data and evaluate its effectiveness in identifying potential outliers. The dataset includes several health indicators including blood pressure, body mass index, waist circumference, glucose, cholesterol, and uric acid. Optimal number of clusters was determined using the Bayesian Information Criterion (BIC). Based on BIC, a four-cluster GMM model provided the most interpretable segmentation. The selected model successfully detected one outlier, which is characterized by a low posterior probability and was most likely caused by a data input error. Compared with Fuzzy C-Means (FCM) and K-Means, GMM had a higher silhouette value, showing clearer and more compact clusters. These findings show the dual capacity of GMM not only for forming interpretable soft clusters but also for identifying anomalous values.