Optimized Quantum Neural Networks for Anomaly Detection in Securities Transactions

Abstract

: In financial systems, abnormal activities in securities transactions pose significant risks due to which there is a requirement of smart and accurate anomaly detection procedures. The presented work in this paper employs a Particle Swarm Optimization (PSO)-tuned Quantum Neural Network (QNN) for anomaly detection. The QNN proposed here is designed using variational quantum circuits to process PSO optimized features and utilises the quantum advantage in neural computing. Performance of the proposed system is evaluated by  executing on the IEEE-CIS Fraud Detection dataset. QNN’s weights and biases are optimized by using PSO to enhance capability of the designed network to escape from local optima and to enhance prediction accuracy. The results presented demonstrate the performance of proposed PSO-QNN model and these outcomes are compared with the conventional algorithms- Long Short-Term Memory (LSTM), k-Nearest Neighbours (KNN), and Support Vector Machines (SVM). The work proposed in this paper achieves a precision of 98.55% with improved sensitivity, specificity, recall and attained an impressive Matthews Correlation Coefficient (MCC) of 0.94. As presented in the results, reduced values of FP and FN rates, showcase the reliability of the proposed PSO-QNN model for anomaly detection. For anomaly detection, adopting QNN and optimising it by PSO, represents a significant advancement and this offers a progressive and functional solution for securities transactions in financial systems.