Clustering Algorithm Based on Sparse Feature Vector for Interval-Scaled Variables

A two-step algorithm, clustering algorithm based on sparse feature vector for interval-scaled variables (CABOSFV_I), is proposed for high dimensional sparse data clustering in this paper. It decomposes a high dimensional problem into several low dimensional ones in first step and then gets the final clusters by second clustering. Because the irrelevant attributes are removed from each cluster in first step, it diminishes the dimensions effectively. Furthermore, the algorithm compresses data effectively by using ´Sparse Feature Vector´. Data scale is reduced enormously, but clustering quality is not affected. Because of the effective dimension deduction and data compression, the algorithm finds clusters in high dimensional large datasets effectively and efficiently.