Research and Application of Improved K-Means Algorithm Based on Fuzzy Feature Selection

K-means is a widely-used clustering algorithm in data mining. In traditional algorithm, each feature is treated equally and each one gives the same contribution to K-means. In fact, redundant and irrelevant features may disturb the clustering result. This paper proposes a improved K-means algorithm based on a fuzzy feature selection strategy. The method is based on measuring ´feature important factor´ (FIF). Firstly, make use of the first time clustering result to get class labels; secondly, set up decision tree to calculate the FIF; thirdly, do the cluster algorithm again with the FIF to modify the similarity measure and then get the modified clustering result. The experiment with UCI datasets proves that, the strategy of fuzzy feature selection can improve the clustering result effectively. At last, the application is done in human resource dataset of a domestic university for further proof of the effectiveness and practicability of the algorithm.