Mixed Quantum–Classical Method for Fraud Detection With Quantum Feature Selection

This paper presents a first end-to-end application of Quantum Support Vector Machine (QSVM) algorithm for classification problem in the financial payment industry using IBM Safer Payments and Computers via Qiskit software stack. Based on real card data, thorough comparison is performed to assess complementary impact brought by current state-of-the-art Learning algorithms with respect Classical Approach. A new method search best features explored Machine's feature map characteristics. The results are compared fraud specific key performance indicators: Accuracy, Recall, False Positive Rate, extracted from analyses based human expertise (rule decisions), classical machine learning (Random Forest, XGBoost) quantum QSVM. In addition, hybrid classical-quantum approach an ensemble model that combines better improve prevention decision. We found, as expected, highly depend selections used select them. QSVM provides exploration space which led improved accuracy mixed quantum-classical detection, drastically reduced data set fit state Hardware.