Improved Clustering Approach based on Fuzzy Feature Selection

Clustering is one of the most heated topics in data mining research. In traditional clustering algorithms, each feature is treated equally and each one does the same contribution to clustering. As a matter of fact, redundant and unrelated features may disturb the clustering result. This paper proposed a fuzzy feature selection strategy to improve the clustering algorithm. The strategy is based on measuring ´Feature Important Factor´ (FIF) to describe the contribution of each feature to the clustering, and makes use of the FIF to get the generalized weight of the contribution of each feature to clustering. In this strategy, the FIF and clustering result are iteratively modified until the result is stable, for the purpose of improving the clustering result. The experiment of K-means algorithm proves that, the strategy of fuzzy feature selection proposed by this paper, can improve the clustering result effectively.