Balanced Spectral Clustering Algorithm Based on Feature Selection

High dimensional data clustering faced some problems such as sparse samples, difficulty in calculating similarity and so on. In addition, the clustering results sometimes be extremely unbalanced with too many or too few category samples. Therefore, we propose a novel algorithm, that is a balanced spectral clustering algorithm based on feature selection. Firstly, the least square method is used to calculate the target loss error. Secondly, the method of feature selection is used to reduce the influence of noise and redundant features. Thirdly, a balanced regularization term exclusive lasso is introduced to balance the clustering results. Finally, the locality preserving projection is used to maintain the feature structure of the samples. A large number of experimental results show that the proposed algorithm outperformed the comparison algorithms on the two indicators (accuracy and normal mutual information) in most cases, which proves the effectiveness of the proposed spectral clustering algorithm.