Affiliations:
School of Mathematics, Harbin Institute of Technology, Harbin, China
Correspondence:
[*]
Corresponding author. Boping Tian, School of Mathematics, Harbin Institute of Technology, Harbin 150001, China. E-mail: [email protected].
Abstract: The fraud problem has drastically increased with the rapid growth of online lending. Since loan applications, approvals and disbursements are operated online, deceptive borrowers are prone to conceal or falsify information to maliciously obtain loans, while lenders have difficulty in identifying fraud without direct contacts and lack binding force on customers’ loan performance, which results in the frequent occurrence of fraud events. Therefore, it is significant for financial institutions to apply valuable data and competitive technologies for fraud detection to reduce financial losses from loan scams. This paper combines the advantages of statistical methods and ensemble learning algorithms to design the grouped trees and weighted ensemble algorithm (GTWE), and establishes fraud prediction models for online loans based on mobile application usage behaviors(App behaviors) by logistic regression, extreme gradient boosting (XGBoost), long short-term memory (LSTM) and the GTWE algorithm, respectively. The experimental results show that the fraud prediction model based on the GTWE algorithm achieves outstanding classification effect and stability with satisfactory interpretability. Meanwhile, the fraud probability of customers detected by the fraud prediction model is as high as 84.19%, which indicates that App behaviors have a considerable impact on predicting fraud in online loan application.
Keywords: Fraud prediction, mobile application usage behaviors, extreme gradient boosting, long short-term memory, grouped trees, weighted ensemble algorithm
