Machine Learning-Based Credit Risk Assessment For Intelligent Lending and Default Prediction in Financial Services

Abstract

A convolutional neural network (CNN) model is proposed to predict credit default in an online lending service environment using public data (e.g., application and user profile data). Given that CNNs have the ability to learn spatial representation, the 1D convolutional layer embedded in the CNN classifier can effectively capture local features along the feature sequence. CNNs can expose correlated features to deep networks for joint learning and enhance performance. Compared with traditional approaches, such as logistic regression or support vector machines, the CNN shows better predictive ability. The effectiveness of deep learning techniques for credit-default prediction from user perspectives presents new insights for intelligent lending services. Despite significant advances in machine learning techniques and their successful applications in many fields, credit-risk-prediction models are still predominantly based on traditional methods. One possible reason is the high-dimensional sparse feature distribution. However, the underlying data sources of credit-risk prediction in the lending industry contribute to high-quality dense representation. An online lending service provides comprehensive application attributes and sufficient historical samples for predictive modeling. With detailed user profiles, the learned model from public data can assist credit-risk management in intelligent lending service. Specifically, a CNN model is proposed to predict credit default in an online lending service environment using public data.