![IconGems-Print](images/IconGems-Print.png)
432 – New Approaches in Classification Methods
Multi-Class ROC Tree and Random Forest for Imbalanced Data Classification
Jiaju Yan
SUNY Stony Brook
Bowen Song
Ocean University of China
Wei Zhu
SUNY at Stony Brook
The imbalanced class problem in classification is highly relevant in many practical scenarios such as the detection of a rare condition. One solution to this problem is to design specific algorithms incorporating the imbalanced classes in the training process of a classifier. In this paper, we propose a multi-class classification tree based on the area under the ROC curve (AUC) to resolve the imbalanced classification problem. This tree classifier aims to maximize the sum of AUC for all one versus all classifiers at the node attribute selection stage, and maximize the harmonic mean of sensitivity and specificity of all one versus all classifiers at the node threshold selection stage. The random forest framework is further applied on the ROC tree with suitable modifications. Volume Under Surface (VUS), the extension of AUC for multiple classes, is discussed in this paper and used to measure the performance of classifiers. The simulation results show that this ROC tree/forest method is superior to CART, random forest and SVM on imbalanced classification problems, while the ROC random forest performs equally well as the usual random forest and SVM on balanced classification problems.