Spectral Clustering Approach with K-Nearest Neighbor and Weighted Mahalanobis Distance for Data Mining

This paper proposes a spectral clustering method using k-means and weighted Mahalanobis distance (Referred to as MDLSC) enhance the degree of correlation between data points improve accuracy Laplacian matrix eigenvectors. First, we used coefficient weight calculate any two constructed set; then, based on matrix, K-nearest neighborhood (KNN) algorithm construct similarity matrix. Secondly, regularized was calculated according normalized decomposed, feature space for obtained. fully considered linear special spatial structure achieved accurate clustering. Finally, various algorithms were conduct multi-angle comparative experiments artificial UCI sets. The experimental results show that MDLSC has certain advantages in each index quality is better. distribution eigenvectors also by more reasonable, calculation maximizes retention characteristics original data, thereby improving algorithm.