A New Approach for Imbalanced Data Classification Based on Minimize Loss Learning

Abstract

The class imbalance problem occurs when instances in one class are more than that in another. It has been reported to severely hinder classification performance of many traditional classification algorithms and many researchers have paid a great deal of attention to this field. Different kinds of methods have been pro-posed to solve the problem these years, such as resampling methods, integrated learning method. However, these conventional class imbalance handling methods might suffer from the loss of potentially useful information, unexpected mistakes or increasing the likelihood of overfitting because they may alter the original data distribution. In this study, we propose a new method for imbalanced data sets which is different from previously proposed solutions to the class imbalance problem. We put forward the idea that treat the performance measures as training target, then designed the loss function and build a model based on artificial neural network to solve the problem. The experimental results on 8 imbalanced data sets show that our proposed method is usually superior to the conventional imbalanced data handling methods.