Threshold optimization and random undersampling for imbalanced credit card data

Abstract Output thresholding is well-suited for addressing class imbalance, since the technique does not increase dataset size, run risk of discarding important instances, or modify an existing learner. Through use Credit Card Fraud Detection Dataset, this study proposes a threshold optimization approach that factors in constraint True Positive Rate (TPR) ≥ Negative (TNR). Our findings indicate Area Under Precision–Recall Curve (AUPRC) score associated with improvement threshold-based classification scores, while positive prior probability causes optimal thresholds to increase. In addition, we discovered best overall results selection are obtained without Random Undersampling (RUS). Furthermore, exception AUPRC, established default yields good performance scores at balanced ratio. evaluation four techniques, eight threshold-dependent metrics, and two threshold-agnostic metrics defines uniqueness research.