A novel technique for fast determination of K in partitioning cluster analysis

Abstract

The input argument k refers to the number of clusters is needed to start all of the probabilistic and possibilistic partitioning algorithms. Although some progress has been made toward its solution, determining this user-specified argument is still one of the main issues in partitioning cluster analysis. Therefore, fast and even automated techniques are needed for determining k in partitioning clustering. In this paper, for determination of k, we proposed the KPEAKS, a simple and fast technique based on the descriptive statistics of peak counts of the features for clustering multidimensional datasets. The experiments on the synthetic and real datasets revealed that the mean of the largest two peak counts and the mean of third quartile and maximum peak count of the features can be successfully used for the estimates of k.