Clustering Algorithm for a Healthcare Dataset Using Silhouette Score Value

Abstract

The huge amount of healthcare data, coupled with the need for data analysis tools has made data mining interesting research areas. Data mining tools and techniques help to discover and understand hidden patterns in a dataset which may not be possible by mainly visualization of the data. Selecting appropriate clustering method and optimal number of clusters in healthcare data can be confusing and difficult most times. Presently, a large number of clustering algorithms are available for clustering healthcare data, but it is very difficult for people with little knowledge of data mining to choose suitable clustering algorithms. This paper aims to analyze clustering techniques using healthcare dataset, in order to determine suitable algorithms which can bring the optimized group clusters. Performances of two clustering algorithms (K-means and DBSCAN) were compared using Silhouette score values. Firstly, we analyzed K-means algorithm using different number of clusters (K) and different distance metrics. Secondly, we analyzed DBSCAN algorithm using different minimum number of points required to form a cluster (minPts) and different distance metrics. The experimental result indicates that both K-means and DBSCAN algorithms have strong intra-cluster cohesion and inter-cluster separation. Based on the analysis, K-means algorithm performed better compare to DBSCAN algorithm in terms of clustering accuracy and execution time.