A multi-view OVA model based on decision tree for multi-classification tasks

Decision tree is a simple classification algorithm and has been widely used in knowledge discovery and pattern recognition fields, which can be used to deal with the multi-classification tasks. In this paper, we present a multi-view OVA model based on decision tree (MVDT) for multi-classification tasks to simplify the structure of the decision tree and improve the generalization ability. A multi-class classification task is divided into c multiple parallel sub-tasks, and MVDT builds c decision trees as base binary classifiers for each sub-task. Each decision tree gives membership vector for each leaf node to estimate the probabilities of the instances in the leaf node belonging to negative classes, as well as presents a precise classification for positive class. Thus, one can obtain more information about instances belonging to negative classes through membership vectors, which helps to achieve higher accuracy and better robustness for classification. As a general framework, MVDT algorithm can use any existing decision tree model as base classifier. To evaluate the performance of our algorithm, we choose C4.5, CART, TEIM, SCDT and NBTree as base classifiers in MVDT. The experiments on 22 data sets show that the proposed MVDT has excellent performance for multi-class classification problems and has excellent robustness to output noise.