Comparison of metrics for feature selection in imbalanced text classification

► Feature selections using Type-I metrics (χP2 and Gini index) achieve the comparable classification performances with those of the combination framework using Type-III metrics (signed χ2 and signed information gain). ► The performances with Type-II metrics (χ2 and information gain) are significantly degraded with increasing the degree of class imbalance. ► Type-III metrics produced the best performance; however, the optimization with these metrics is not easy in real applications. ► Type-I metrics serve as more simplified alternative methods for the combination framework. ► The classification performances using Type-I and Type-II metrics have positive correlations with the number of negative features.