Abstract: Feature selection is an important step during data mining and machine learning. With the lack of labeled instances, the problem of effective selection is worthy of consideration. This paper proposed a novel semi-supervised fealure selection algorithm based on the definition of inter-set and infra-set correlation,which starts from the original and small labeled samples and gains the final clusters by extension of labels. A complex evaluation was utilized as criterion to find optimal feature subset. Finally, the experimental results demonstrate the efficacy of the algorithm.

Key words: Feature selection, Semi-supervised, Inter-set correlation, Intra-set correlation