Anomaly detection framework using enhanced isolation forest for GDPR-DPIA-compliant credit card fraud detection in e-commerce

The rapid growth of e-commerce has increased credit card fraud while simultaneously increasing concerns regarding data privacy, accountability, and regulatory compliance. Fraud detection in this domain remains challenging due to extreme class imbalance, evolving fraud patterns, and limited transparency in many machine learning models. This study proposes an enhanced isolation forest (EIF) framework that integrates hybrid normalization, Pearson correlation-based feature selection, and hyperparameter optimization using RandomizedSearchCV. Experiments were conducted on the Kaggle–ULB credit card fraud detection dataset, which contains anonymized European cardholder transactions and represents a highly imbalanced real-world setting. The proposed EIF framework improves fraud detection performance while preserving data imbalance, achieving 99.8% accuracy, 99.9% precision, 99.7% recall, 99.8% F1-score, and AUROC 0.98, outperforming baseline unsupervised anomaly detection models. The study contributes a statistically robust and explainable anomaly detection framework that integrates optimization strategies with General Data Protection Regulation (GDPR) and Data Protection Impact Assessment (DPIA) aligned data governance, supporting reliable and compliant fraud detection in intelligent e-commerce systems.