Imbalanced Data Classification Performance Evaluation: AUC and G-mean Metrics

Experimental Setup\nThis study uses 10 imbalanced binary classification datasets from the KEEL dataset. Table 1 shows the size, feature size, and imbalance ratio of these 10 datasets. The imbalance ratios of the selected datasets range from 1.87 to 72.69, which can effectively reflect the performance of our method on highly imbalanced and slightly imbalanced datasets. The experiment uses two-fold cross-validation, and the average of 100 experiments is used as the result.\n\nExperimental Metrics\nThis study uses two metrics to evaluate the classification performance of the proposed algorithm on imbalanced datasets. AUC is commonly used as a metric for imbalanced data classification, and it represents the area under the ROC curve. G_mean is also a commonly used evaluation metric for imbalanced classification, which can be represented as.