Recall and Selectivity Normalized in Class Labels as a Classification Performance Metric

In the supervised classification approach, a significant research issue is the selection of a performance metric suitable for an imbalanced dataset. So far, many classification performance metrics have been defined and analyzed from different perspectives regarding the imbalanced dataset. However, there is still no clear indication of which metric is universal and can be used for any skewed data problem. This study introduces a new classification performance metric based on the harmonic mean of recall and selectivity normalized in class labels. The results show that the proposed performance metric is robust. Thus, the proposed performance metric is significantly less sensitive to changes in the majority class and more sensitive to changes in the minority class than other existing classification performance metrics. It is proven that the proposed classification performance metric has the same value as other evaluation metrics in the case of equal recall and selectivity. Additionally, the properties of the proposed performance measure metric are analyzed using the ROC isometric space.