EFFECTIVE EVALUATION OF CLUSTERINGAPPROACHES FOR MINING MEDICAL DATASETS

Abstract

With the explosion of medical data in recent years, there’s a
growing need for effective data mining techniques to uncover patterns that can enhance clinical decision-making. Among these techniques, clustering—an unsupervised learning method—has proven especially useful for discovering natural groupings in data without needing pre-labeled outcomes. This study explores and compares the performance of four clustering algorithms—K-Means, Hierarchical Agglomerative Clustering, DBSCAN, and Fuzzy C-Means—across three widely recognized medical datasets: Diabetes, Hepatitis, and Breast Cancer. To assess their effectiveness, both internal validation measures and external metrics are discussed for each algorithm as the performance varies depending on the characteristics of the dataset. For instance, DBSCAN excelled in identifying complex patterns and filtering out noise, particularly in the Breast Cancer dataset.
Meanwhile, Fuzzy C-Means showed strength in handling overlapping clusters within the Diabetes dataset. These findings provide useful insights for researchers and healthcare professionals looking to apply the right clustering methods in medical data analysis.