Generative adversarial network based telecom fraud detection at the receiving bank.

Area of Science:

• Financial Technology
• Machine Learning
• Cybersecurity

Background:

• Telecom fraud poses a significant challenge, particularly in developing nations like China.
• Coordinating inter-agency efforts to combat fraud remains difficult.
• Detecting large fund transfers at receiving banks is crucial for fraud prevention.

Purpose of the Study:

• To develop a novel generative adversarial network (GAN) based model for detecting fraudulent large money transfers.
• To assign a fraud probability score to each large transfer for timely intervention.
• To enhance the security measures of financial institutions against telecom fraud.

Main Methods:

• Utilized a deep denoising autoencoder to model complex probabilistic relationships in transaction data.
• Employed adversarial training with a generator and discriminator in a minimax game.
• Developed a GAN-based model to calculate the likelihood of a transfer being fraudulent.

Main Results:

• The proposed GAN model demonstrated superior performance compared to established classification methods.
• Experimental results showed the model's effectiveness in real-world banking applications.
• Achieved a reduction in financial losses of approximately 10 million RMB over twelve weeks in two commercial banks.

Conclusions:

• The GAN-based model offers an effective solution for detecting large fraudulent transfers.
• Successful implementation in commercial banks led to significant financial savings and reputational enhancement.
• This approach provides a robust mechanism for financial institutions to mitigate telecom fraud risks.