A Deep Learning Method of Credit Card Fraud Detection Based on Continuous-Coupled Neural Networks

Yanxi Wu, Liping Wang, Hongyu Li, Jizhao LiuMathematics 2025 · DOI

TL;DR

This paper introduces a novel credit card fraud detection system using a brain-inspired Continuous-Coupled Neural Network (CCNN). By mimicking the dynamic and complex processing of the human brain, the CCNN, combined with data balancing techniques, achieves near-perfect accuracy in identifying fraudulent transactions, outperforming traditional machine learning methods.

What Problem Does This Paper Address?

Credit card fraud is a massive and growing problem in the financial sector, costing billions annually. As payment technologies evolve, so do fraud tactics, making detection increasingly difficult. While machine learning has been applied, deep learning methods are still underutilized, especially in handling the inherent class imbalance (vastly more legitimate transactions than fraudulent ones) and the complex, evolving nature of fraud patterns. Existing methods often struggle to accurately identify rare fraudulent activities without being overwhelmed by the majority of normal transactions.

Methods

The core of this work is the Continuous-Coupled Neural Network (CCNN), a brain-inspired model designed to better capture complex temporal and spatial patterns. Unlike traditional neural networks, the CCNN uses continuous neuron activation and dynamic coupling mechanisms, aiming to mimic the brain's ability to process information more dynamically.

To tackle the significant class imbalance in credit card transaction data, the researchers employed the Synthetic Minority Oversampling Technique (SMOTE). SMOTE generates synthetic minority class samples (fraudulent transactions) by interpolating between existing minority samples and their neighbors, effectively balancing the dataset.

For model input, the traditional feature vectors were transformed into a single-channel matrix, essentially treating each transaction as a "mini-image." This allows the CCNN to leverage its feature extraction capabilities, similar to image processing tasks. The CCNN layer is then followed by a fully connected layer for final classification.

Datasets

The primary dataset used in this study is the Kaggle Credit Card Fraud Detection (CCFD) dataset. This dataset contains 284,807 transactions with 30 features, including time, anonymized principal components (V1-V28), and transaction amount. It's noted for its severe class imbalance, with only 0.17% of transactions being fraudulent. The paper does not introduce new benchmarks but uses this publicly available dataset for evaluation.

Key Insights and Ideas

  • Brain-like Computing for Finance: The paper demonstrates that principles from brain-like computing, specifically the CCNN, can be effectively applied to solve complex financial problems like fraud detection, opening new avenues beyond traditional ML.
  • Dynamic Pattern Recognition: The CCNN's ability to handle dynamic and chaotic patterns is particularly well-suited for fraud detection, where fraudulent behaviors constantly evolve to evade detection.
  • SMOTE's Crucial Role: The successful application of SMOTE highlights its importance in making imbalanced datasets tractable for deep learning models, significantly improving the detection of minority class (fraudulent) instances.
  • Matrix Transformation for Visual Models: Transforming tabular data into a matrix format for a "visual" model architecture is an interesting approach that allows leveraging CNN-like feature extraction capabilities on non-image data.

Key Results

The CCNN model, particularly with a time step (T) of 2, achieved exceptional performance on the Kaggle CCFD dataset:

  • Accuracy: 0.9998
  • Precision: 0.9996
  • Recall: 1.0000
  • F1-Score: 0.9998

These results significantly outperform traditional machine learning models like Decision Trees, KNNs, and Logistic Regression, and are competitive with or surpass other advanced deep learning methods mentioned. The model also demonstrated high efficiency, with a training time of approximately 32 seconds per epoch on a high-end GPU, making it suitable for real-time applications.

Limitations and Future Work

The authors acknowledge several limitations and propose future directions:

  • SMOTE Limitations: The interpolation used by SMOTE might not fully capture the complexity of real fraudulent transactions, potentially leading the model to fit synthetic data. Future work aims to explore alternative resampling methods that balance the dataset without over-fitting to synthetic samples.
  • Hyperparameter Optimization: Finding the optimal hyperparameters for the CCNN was time-consuming. Future research will involve a more systematic grid search with a smaller step size.
  • Dataset Scope: The current study is limited to the Kaggle CCFD dataset. Future work plans to validate the CCNN model on a wider range of credit card fraud datasets to assess its generalization capabilities. The authors also note the challenge of obtaining diverse, real-world transaction data due to privacy concerns.

Summary updated

Abstract

With the widespread use of credit cards in online and offline transactions, credit card fraud has become a significant challenge in the financial sector. The rapid advancement of payment technologies has led to increasingly sophisticated fraud techniques, necessitating more effective detection methods. While machine learning has been extensively applied in fraud detection, the application of deep learning methods remains relatively limited. Inspired by brain-like computing, this work employs the Continuous-Coupled Neural Network (CCNN) for credit card fraud detection. Unlike traditional neural networks, the CCNN enhances the representation of complex temporal and spatial patterns through continuous neuron activation and dynamic coupling mechanisms. Using the Kaggle Credit Card Fraud Detection (CCFD) dataset, we mitigate data imbalance via the Synthetic Minority Oversampling Technique (SMOTE) and transform sample feature vectors into matrices for training. Experimental results show that our method achieves an accuracy of 0.9998, precision of 0.9996, recall of 1.0000, and an F1-score of 0.9998, surpassing traditional machine learning models, which highlight CCNN’s potential to enhance the security and efficiency of fraud detection in the financial industry.
BibTeX
@inproceedings{wu2025deep,
  author = {Yanxi Wu and Liping Wang and Hongyu Li and Jizhao Liu},
  title = {{A Deep Learning Method of Credit Card Fraud Detection Based on Continuous-Coupled Neural Networks}},
  booktitle = {Mathematics},
  year = {2025},
  doi = {10.3390/math13050819},
}
Figure