A Graph Coloring Based Framework for Views Construction in Multi-View Ensemble Learning

Multi-view ensemble learning (MEL) is new and fast growing area of machine learning. Here, the subsets of multiple features of same dataset are used for the learning and their prediction is finally ensembled. Vertical partition of dataset corresponds to the subset of the feature set is considered as view in a single-sourced dataset. View construction is an important task in MEL because the quantity of views and their quality leads to better performance of MEL. There is a well known graph partitioning technique, called graph coloring that assign the vertices with different color so that no two adjacent vertices have same color. This technique has various applications in different fields. The one of the applications of coloring is in clustering. In this paper, graph coloring is used for the creation of views in multiview ensemble learning. Here uncorrelated (dissimilar) feature clustering is done with the help of coloring. This paper presents a Graph coloring based views creation (GC-VC) method for automatic views creation in multi-view ensemble learning(MEL). To illustrate the performance of the framework, the five different classifiers have been employed namely Support Vector Machine (SVM), K-nearest neighbors (KNN), Naive Baysian (NB), Neural Network (NN), Decision Tree (DT). The experiments have been performed on ten different datasets from the UCI repository. The results and their non-parametric statistical analysis of MEL have been done. A good enhanced classification accuracy has been achieved by the purposed MEL framework.