Open AccessA Loan risk assessment model with consumption features for online finance
Author(s) -
Li Li,
Ding Huang
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1848/1/012068
Subject(s) - loan , computer science , construct (python library) , consumption (sociology) , cluster analysis , credit risk , amortizing loan , convolutional neural network , artificial intelligence , machine learning , data mining , finance , business , non performing loan , bridge loan , computer network , social science , sociology
While online finance is rapidly growing, it is a big challenge to evaluate the loan risk of users based on the data on the Internet. In this study, we used basic user information and added the user’s consumption features to construct a loan risk assessment model that integrates features. We extract consumption features from two aspects: firstly, construct consumption portrait features through statistical analysis and clustering; secondly, combine convolutional neural networks (CNN) and bidirectional long short-term memory (BiLSTM) networks to extract sequence features in consumption, and add attention mechanisms to improve the evaluation effect. Finally, the features are combined and fed into the fully connected layer, and the probability of default loan is calculated through the activation function. Our data set comes from an online finance company. The experimental results show that the loan assessment model combined with consumption features does improve the accuracy of default loans under the AUC and KS indicators.