K-Means Clustering for Liver Patient Stratification Using Elbow Method and Silhouette Score Validation

Abstract

Abstract—Liver disease is one of the global health problems that causes millions of deaths every year, with a high incidence rate in India due to hepatitis infection, alcohol consumption, and unhealthy lifestyle factors. The treatment of liver disease requires accurate early diagnosis, but manual identification of clinical patterns in patients takes time and is prone to interpretation errors. Various studies have utilized biochemical data from patients for liver disease classification, but automatically grouping patients based on the similarity of biochemical profiles without diagnostic labels is still uncommon. This study applies the K-Means algorithm on the Indian Liver Patient Dataset containing 583 patient medical records with 10 biochemical and demographic features, combined with preprocessing including missing value imputation, label encoding, and StandardScaler normalization, as well as the use of the Elbow Method to determine the optimal number of clusters and the Silhouette Score for cluster quality evaluation. The results show that k=3 is the best number of clusters with a Silhouette Score of 0.208, producing three patient groups, namely critical condition with very high bilirubin and enzyme levels, normal biochemical condition, and moderate biochemical disorder. These findings indicate that K-Means has the potential to be used as an early patient stratification tool for liver disease to support the development of artificial intelligence-based clinical decision support systems.