Fraudulent Credit Card Transaction Monitoring Using Deep Learning: A CNN-LSTM Approach for Single-Transaction Security Identification

Credit card cash-out methods have become increasingly complex, with new fraudulent transaction forms emerging continuously. Effective management is hindered by challenges in obtaining accurate transaction data during customer interactions, relying heavily on account-level cash-out label data. This study proposes a deep learning-based single-transaction security identification model using Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. A Spy mechanism is introduced for labeling sample data, extracting 1.2 million highly reliable positive samples and labeling 1.5 million negative samples from transaction data. The model leverages 50 key variables selected from 148 candidates, including customer attributes, credit utilization, and transaction preferences. The proposed CNN-LSTM hybrid model achieves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6 \%}$</tex> accuracy in detecting fraudulent transactions, with a Population Stability Index (PSI) of 0.05 %, demonstrating its robust monitoring capabilities. This research provides an innovative method for enhancing transaction security in commercial bank credit card systems, addressing challenges in obtaining high-accuracy labeled data and improving the discriminative performance of deep learning algorithms.