Application of a density based clustering technique on biomedical datasets

The detection of the number of clusters in a biomedical dataset is very important for generating high quality clusters from the biomedical dataset. In this paper, we aim to evaluate the performance of a density based K-Means clustering technique called DenClust on biomedical datasets. DenClust produces the number of clusters and the high quality initial seeds from a dataset through a density based seed selection approach without requiring an user input on the number of clusters and the radius of the clusters. The high quality initial seeds for K-Means results in high quality clusters from a dataset. The performance of DenClust is compared with six other existing clustering techniques namely CRUDAW-F, CRUDAW-H, AGCUK, GAGR, K-Means, and K-Means++ on the twenty biomedical datasets in terms of two external cluster evaluation criteria namely Entropy and Purity and one internal cluster evaluation criteria called Sum of Squared Error (SSE). We also perform a statistical non-parametric sign test on the cluster evaluation results of the techniques. Both the cluster evaluation results and statistical non-parametric sign test results indicate the superiority of DenClust over the existing techniques on the biomedical datasets. The complexity of DenClust is O(n2) but the overall execution time of DenClust on the datasets is less than the execution time of AGCUK and GAGR having O(n) complexity.