Highly imbalanced classification using improved rotation forests
Xiaonan Fang, Xiyuan Zheng, Yanyan Tan, Huaxiang Zhang · International Journal of Wireless and Mobile Computing · 2016
Imbalanced data classification is a challenging problem in data mining. It happens in many real-world applications and has attracted growing attentions from researchers. This issue occurs when the number of one class is much higher than the other class. Ensemble of classifiers has been well known as an effective solution. Then, two novel ensemble algorithms (RUROForest and SROForest) based on rotation forests are proposed for solving highly imbalanced problems. Random under-sampling or SMOTE approaches are combined with rotation forest in the proposed algorithms, which balance the uneven distribution of data sets and keep the diversity of single classifier as well. Focused on two-class highly imbalanced problems, 22 relevant data sets are performed in experiments. Experimental results and statistical analyses show that our proposed methods overtake the state-of-the-art ensemble methods on the most widely used imbalanced measure criterion AUC.