Clustering analysis using an adaptive fused distance

The selection of a proper distance function is crucial for analyzing the data efficiently. To find an appropriate distance for clustering algorithm is an unsolved problem as of now. The purpose of this study is to introduce an adaptive fused distance. The S-distance is integrated with the Euclidean distance with the help of a statistical coefficient that depends on density variance of a dataset. We afterward propose a modified kappa-means clustering algorithm using the novel distance in order to achieve improvement in clustering by finding out the natural and obscure patterns in the data. Some useful properties of the novel distance metrics are elaborated. Theoretical convergence analysis of the proposed clustering is addressed. All the experiments are performed on fourteen datasets. Empirical results using five clustering evaluation metrics on fourteen datasets illustrate that the proposed clustering algorithm defeats seven state-of-the-art clustering methods before and after adding noisy features. It is also proved that the proposed clustering algorithm is statistically significant.