Hybrid stacking and logistic regression meta-model for credit card fraud detection using SMOTE and multiple machine learning algorithms
Abstract This study presents a stacking ensemble framework for credit card fraud detection on the imbalanced Kaggle dataset, using SMOTE to address class imbalance. Base classifiers (RF, SVM, KNN, NB, GB, LR) are evaluated with accuracy, precision, recall, and F1-score, with RF and KNN emerging as the strongest learners, reaching 97% recall and 0.98 F1-score after SMOTE. The final ensemble employs RF and KNN as base learners with Logistic Regression as the meta-classifier, and pairwise prediction correlations (r ≈ 0.34 for RF–KNN vs. r ≈ 0.81 for RF–GB) confirm that RF and KNN form a more diverse and complementary pair. A refined stacked hybrid using RF–KNN–GB with meta-features and original data achieves the best trade-off for fraud and intrusion detection, obtaining 0.975 recall with only one false positive, 0.9968 precision, 0.986 F1-score, and accuracy of 0.9997, outperforming existing hybrid methods. These findings demonstrate that carefully designed ensembles and feature engineering can deliver robust fraud detection on highly imbalanced data, surpassing individual models and reported
Authors
- Haitham A. Ghalwash (ORCID: https://orcid.org/0000-0001-8887-2716)
- Mohamed M. Abbassy (ORCID: https://orcid.org/0000-0003-2069-6773)
Institutions
- Beni-Suef University (EG)
Publication Details
- Journal
- Scientific Reports
- Published
- 2026-11-13
- DOI
- https://doi.org/10.1038/s41598-026-73942-5
- Primary Topic
- Imbalanced Data Classification Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- Science and Technology Development Fund