Multi-view Ensemble Learning

Abstract

Partitioning the feature set into non-empty subsets of features is the generalized task of feature subset selection. The subsets of features are collectively useful than a subset of the feature. The composition of classification models of the multiple views corresponding each subset of the feature-set can outperform a single-view classifier. Multi-view ensemble learning (MEL) exploits the views of the dataset to enhance the classification using consensus and complementary information. The way of partitioning of feature set affects the classification performance of MEL. Therefore, supervised feature set partitioning (SFSP) method is proposed and compared with random feature set partition (RFSP) method. Experiments have been performed on seven high dimensional datasets. The results and their statistical analysis show that SFSP method is better than RFSP method for MEL.