A Classification Algorithm Based on Ensemble Feature Selections for Imbalanced-Class Dataset
Hua Yin, Keke Gai, Zhijian Wang · 2016
Traditional classification algorithms addressing imbalanced-class dataset mostly concentrate on the majority classes' accuracy, such that the minority class's accuracy is usually ignored. Focusing on this issue, we propose a novel classification algorithm using Ensemble Feature Selections (EFS) for imbalanced-class dataset. This algorithm utilizes the superiority of EFS in accuracy, then considers the diversity and imbalance in the designing appropriate feature subset objective function to make it fit for the imbalanced dataset. It chooses the minority class-oriented F measurement for computing accuracy and imports a punishment-reward mechanism into the KW diversity measurement. When the minority class's accuracy goes up, the reward-factor is given. Otherwise, the punishment-factor is given. Comparing with four algorithms, our experimental evaluations have showed that Mostly our algorithm can improve the accuracy of minority class.