Research on a financial fraud identification model by fusing a convolutional neural network

Area of Science:

• * Financial analysis and machine learning applications.
• * Corporate governance and financial risk management.

Background:

• * Existing financial fraud identification methods for listed companies suffer from low accuracy and poor real-time performance.
• * Detecting financial fraud is crucial for maintaining market integrity and investor confidence.

Purpose of the Study:

• * To propose a novel hybrid identification model, CNN-SVM, to overcome the limitations of current financial fraud detection techniques.
• * To enhance the accuracy and real-time capabilities of identifying fraudulent activities in listed companies.

Main Methods:

• * A hybrid model fusing a Convolutional Neural Network (CNN) for feature extraction and a Support Vector Machine (SVM) for classification.
• * Utilized a dataset of 7,429 samples from non-financial A-share listed companies penalized for fraud (2007-2022).
• * Employed random oversampling to create a balanced training set of 13,540 samples, focusing on 87 indicators across corporate governance, financial oversight, and operational metrics.

Main Results:

• * The CNN-SVM model achieved a significant performance leap with an Area Under the Curve (AUC) of 0.97, recall of 0.99, and F1-score of 0.97.
• * Demonstrated superior performance compared to traditional logistic regression and random forest models.
• * The model effectively extracts high-level features for accurate fraud classification.

Conclusions:

• * The proposed CNN-SVM framework offers a superior tool for real-time financial risk control.
• * The hybrid model effectively balances training efficiency with high precision in financial fraud identification.
• * Findings support the adoption of advanced machine learning techniques for robust corporate financial oversight.