Heterogeneous ensemble feature selection based on weighted Borda count

Feature selection is important step in many data mining applications. Reduction of data dimensionality through feature selection reduces computational time, complexity and provide better interpretability. Besides well established feature selection approaches such as filter, wrapper and embedded approach, novel methodology emerged recently: ensemble feature selection. This approach utilize diversity to select final feature subset. In this paper, we proposed four novel heterogeneous ensemble methods based on eight basal feature selection techniques in first stage and modified Borda count voting schemes in the second stage. The proposed methods were evaluated on four artificial datasets achieving significantly higher index of success than conventional feature selection techniques.