Skip to Main Content
Among various redundancy based gene selection methods, the Fast Correlation-Based Filter (FCBF) is one of the most effective. FCBF works in an iterative way, where one predominant feature is selected at each step and then some redundant features are removed by the selected one. However, the size of selected feature subset is not considered by FCBF, and weakly relevant features are too inclined to be eliminated. Aiming at this problem, this paper proposes a new approximate Markov blanket definition for FCBF, which strengthens the criterion for redundant features. Based on the new definition, the size of the selected feature set is used to adjust the criterion dynamically. Experimental results on several real gene data sets demonstrated the outstanding performance of the proposed algorithm compared with other several state-of-arts techniques.