Using an Optimized Learning Vector Quantization- (LVQ-) Based Neural Network in Accounting Fraud Recognition.

Area of Science:

• Artificial Intelligence
• Financial Risk Management
• Machine Learning

Background:

• Artificial intelligence (AI) and artificial neural networks (ANNs) are increasingly applied to fraud identification.
• Learning Vector Quantization (LVQ) neural networks are prominent in fraud detection due to high accuracy.
• Existing models require optimization for enhanced fraud risk recognition.

Purpose of the Study:

• To explore the application of LVQ neural networks in fraud identification.
• To propose an optimized LVQ-based combined neural network model for fraud risk recognition.
• To evaluate the effectiveness of the proposed model using real-world financial data.

Main Methods:

• Utilized 550 listed companies with fraud (2015-2019) and 550 non-fraud matched companies as samples.
• Selected key fraud risk identification indicators from literature, addressing collinearity via paired sample T-test and principal component analysis.
• Developed and tested an optimized LVQ-based combined neural network model.

Main Results:

• Identified five key indicators with optimal fraud identification effects after rigorous statistical analysis.
• The optimized LVQ-based combined neural network model showed improved fraud risk recognition rates compared to standard models.
• Empirical validation confirmed the model's effectiveness in distinguishing fraudulent from non-fraudulent companies.

Conclusions:

• The optimized LVQ-based combined neural network model offers a significant advancement in corporate fraud risk identification.
• The study highlights the potential of refined AI techniques in combating financial fraud.
• Future research should focus on further developing and validating advanced fraud risk identification models.