Classification via correlation-based feature grouping

Employing the most relevant and discriminating features is very important to achieve a successful classification with low computational cost. Although, different feature selection methods have been recently developed for this purpose, feature grouping can deal with high dimensional sparse feature vectors more effectively, yielding better interpretation of the data. In this paper, a correlation-based feature grouping (CFG) method is proposed. First, the features are grouped based on the variety of their correlation scores, and then, a new representative feature vector is generated for each group by combining its features. To investigate the strength of CFG method, two filter methods of χ² and correlation are employed for feature selection, while classification is performed using a support vector machine (SVM) and k-Nearest Neighbor (k-NN). The empirical study on two datasets of protein-protein interactions (PPIs) and breast cancer verifies that the idea of employing feature grouping is more efficient than employing feature selection in identifying a set of features that exhibit high classification accuracy. In addition, a CFG diagram is introduced in this paper, which is used to visualize the groups and their corresponding features found by the proposed method.