An improved index for clustering validation based on Silhouette index and Calinski-Harabasz index

Abstract

The evaluation of clustering effects has been an important issue for a long time. How to effectively evaluate the clustering results of clustering algorithms is the key to the problem. The clustering effect evaluation is generally divided into internal clustering effect evaluation and external clustering effect evaluation. This paper focuses on the internal clustering effect evaluation, and proposes an improved index based on the Silhouette index and the Calinski-Harabasz index: Peak Weight Index (PWI). PWI combines the characteristics of Silhouette index and Calinski-Harabasz index, and takes the peak value of the two indexes as the impact point and gives appropriate weight within a certain range. Silhouette index and Calinski-Harabasz index will help improve the fluctuation of clustering results in the data set. Through the simulation experiments on four self-built influence data sets and two real data sets, it will prove that the PWI has excellent evaluation of clustering results.