Online Loan Default Prediction Model Based on Deep Learning Neural Network

Area of Science:

• Financial Technology (FinTech)
• Machine Learning in Finance
• Risk Management

Background:

• The rapid growth of Internet lending necessitates robust default prediction models.
• Peer-to-peer (P2P) online lending platforms offer reduced costs and improved capital efficiency for borrowers.
• Integrating platform data with third-party information is key for accurate user default behavior prediction.

Purpose of the Study:

• To analyze the risks and challenges within P2P online lending platforms.
• To introduce the principles and characteristics of the Backpropagation Neural Network (BPNN) model.
• To establish a credit risk rating process for online lending using BPNN.

Main Methods:

• Data cleaning and variable selection on credit customer data from lending clubs.
• Development of an online lending default risk assessment model utilizing BPNN.
• Comparative simulation and performance analysis against Support Vector Machine (SVM) and regression models.

Main Results:

• The BPNN model achieved a highest accuracy rate of 98.01% and a recall rate of 99.82%.
• The Area Under the Curve (AUC) for the BPNN model was 0.79, significantly higher than SVM and regression models.
• The developed DPNN model demonstrated superior performance compared to traditional methods.

Conclusions:

• The DPNN-based online loan default prediction model holds significant practical application value.
• Proactive prediction of customer default risk can mitigate losses for P2P companies and lenders.
• Implementing advanced prediction models promotes stability and competitiveness within the P2P lending industry.