K-Data Depth Based Clustering Algorithm

摘要

This paper proposes a new data clustering algorithm based on data depth. In the proposed algorithm the centroids of the K-clusters are calculated using Maha-lanobis data depth method. The performance of the algorithm called K-Data Depth Based Clustering Algorithm (K-DBCA) is evaluated in R using datasets defined in the mlbench package of R and from UCI Machine Learning Repository, yields good clustering results and is robust to outliers. In addition, it is invariant to affine transformations and it is also tested for face recognition which yields better accuracy.