Optimising Credit Card Fraud Detection: A Validation-Weighted Random Forest and XGBoost Ensemble

Dammu Ramadevi, Shaik Dilnawaz

Abstract: Credit card fraud detection is dominated by a severe class imbalance in which fraudulent records constitute a fraction of one percent of transaction volume, so classifiers tuned for accuracy alone conceal unacceptable false-negative rates. This paper develops a weighted soft-voting ensemble of Random Forest and XGBoost for the public ULB credit card dataset of 284,807 European transactions. Class skew is addressed through balanced subsampling and gradient scaling rather than synthetic oversampling, and both the ensemble weights and the operating threshold are selected on a held-out validation partition to avoid test-set leakage. On the untouched test partition the ensemble attains an F1-score of 0.8804, precision of 0.9419, Matthews correlation of 0.8821, and AUPRC of 0.8695, reducing false alarms below either base learner at identical recall while completing training in 77 s on a single CPU core.

Keywords: Credit Card Fraud Detection, Random Forest, XGBoost, Ensemble Learning, Class Imbalance, Precision-Recall Analysis

How to Cite?: Dammu Ramadevi, Shaik Dilnawaz, "Optimising Credit Card Fraud Detection: A Validation-Weighted Random Forest and XGBoost Ensemble", Volume 15 Issue 7, July 2026, International Journal of Science and Research (IJSR), Pages: 2061-2066, https://www.ijsr.net/getabstract.php?paperid=SR26720110935, DOI: https://dx.doi.org/10.21275/SR26720110935

Download Citation: APA | MLA | BibTeX | EndNote | RefMan