Differential Privacy enabled Privacy-Preserving Neural Network Processing for Credit Risk Prediction

In the modern era of data-driven decision-making, predicting bank loan approvals is pivotal in the financial industry. However, using sensitive customer data for these predictions raises significant privacy concerns. To address these issues, this study proposes a Differential Privacy enabled Privacy-Preserving Neural Network Processing for Credit Risk Prediction, leveraging Laplace noise-based differential Privacy, and compares it with the traditional Gaussian Noise mechanism-based differential privacy. The proposed Neural Network framework focuses on ensuring data privacy at all stages of the bank loan prediction process, including the training and inference phases, with privacy to the training data, input data, the model, and the output data. By maintaining data privacy throughout the entire process, from training, and model development to testing and prediction, this research addresses the critical issue of protecting sensitive customer information while still making accurate bank loan predictions. This comprehensive approach minimizes the risk of privacy breaches and maintains the integrity of the prediction model. Key findings of this research include a comparative assessment of the Laplace noise mechanism against the Gaussian noise mechanism, highlighting the tradeoffs between data utility and privacy preservation. Furthermore, the security analysis demonstrates that the Laplace noise mechanism ensures ε-differential Privacy, safeguarding sensitive customer information while allowing accurate loan approval predictions.