International Journal of Advanced Research in Computer and Communication Engineering

Abstract: This paper proposes and experimentally confirms a ready-for-production hybrid quantum-classical protocol for detection of fraudulent transactions under credit card and UPI payment methods. We create an Integrated Dataset consisting of 3,044,322 anonymized transactions from different public and institutional sources and suggest a modular quantum-inspired feature-engineering flow that generalizes a 14-dimensional raw feature vector to 268 engineered descriptors employing amplitude/phase encodings, entanglement-motivated pairwise interactions, as well as measurement-pro Quantum-inspired features are integrated with efficient classical learning algorithms (LightGBM, XGBoost, CatBoost, RandomForest, ExtraTrees) in a learned stacking ensemble we call quantum weighted ensemble. It was trained with stratified 5-fold cross-validation with SMOTE-aware rebal applied solely on folds to reduce leakage. In a reserved temporal holdout test set, proposed system achieves AUC = 0.9269, Precision = 0.92, Recall = 0.96, F1 = 0.94, and latency 2.7 ms per transaction in simulated production mode. In comparison with a classically tuned RandomForest Baseline (AUC # 0.8851), The hybrid system reduces false positives to as low as ~71.2 false negatives by "~74.8%, attaining estimated yearly savings for a 100M transaction operator of the magnitude of US$1.36M due to lowered loss as well as research expenses. Ablation Scientific research reveals that largest marginal gains are entanglement-related pairwise encodings due to measurement probabilistic descriptions. All results are significant at a 0.001 level (.p < 0.001). required no special quantum hardware (quantum encodings are classically calculatedRoute) but is designed to be portable to NISQ devices for additional improvements. We elaborate on operational limitations, moral implications (confidentiality, transparency, rectification of false positives), and specific next: kernel porting on quantum processors, federated quantum learning for multi-bank cooperative learning and adversarial robustness assessment. This research shows a practicable method for almost- Quantum concepts to concretely enhance financial fraud defenses while remaining executable now.
Keywords: Quantum Machine Learning, Fraud Detection, Hybrid Computing, Financial Analytics.

Downloads: | DOI: 10.17148/IJARCCE.2025.141009