Automatic Neighborhood Search Clustering Algorithm Based on Feature Weighted Density

Abstract

— The failure of traditional clustering methods on high-dimensional data has been a thorny problem. Therefore, we propose a simple but effective mean shift feature weighted deformation method (WDNS) to calculate the density value of high-dimensional data points by learning the weights of the features. The neighborhood search is then carried out using the density center in the decision diagram as the starting point, and the points of the same cluster are merged to finally complete the clustering. The experimental results show that the algorithm has higher clustering accuracy than the six existing clustering algorithms. In addition, it has the outstanding feature of automatic parameter setting, which is not available in its peers. In summary, this work can improve the state-of-the-art of clustering algorithms.