Research And Implementation Of Anti-fraud System For Credit Leasing Business

With the rapid development of the leasing economy,the traditional model with deposit as its core has gradually lost vitality,and the new model of replacing deposit with credit has become the development trend of the industry.In credit leasing,order submitted by user can enjoy the deposit-free leasing service after being reviewed by the credit leasing platform.Since there is no longer deposit in the process to limit the user's breach of contract,the operating cost of frauds such as swindles and scams is lower,and it is easier to become the target of various fraud attacks.Different from general fraud,in the credit leasing scenario,the user's fraud intention is easily affected by people around.If people around swindle and scam successfully,the user will have a stronger tendency to deceive when ordering.Therefore,traditional anti-fraud methods that only consider individual characteristics are difficult to achieve good results in the credit leasing scenario.In addition,with the gradual expansion of the credit leasing market and the audience,the review pressure on credit leasing platform has gradually increased,and it has high requirements for the timeliness of the feedback of review results.This shows that fraud detection in the credit leasing scenario needs to take into account the accuracy and timeliness of the results.Concerning the problem of anti-fraud in credit leasing business,the main contributions of this thesis are as follows:1.Based on user order and behavior data,we construct an order relationship network with multi-attribute edges,which reduces network complexity while retaining network information.2.An edge set based fraud detection algorithm ES-FDA is designed according to the graph neural network framework,taking into account individual characteristics and network structure characteristics.The experimental results on the order relation network show that the ES-FDA algorithm has a significant improvement on F1-score and AUC compared with the existing methods.3.Based on the order relationship network,we design and implement an antifraud system that can receive streaming data and make real-time predictions.Test results show that the system meets the functional and performance requirements of the anti-fraud system in the credit leasing scenario.