Transfer Learning Method Based On Tri-training And Its Application In Credit Field

In recent years,supervised learning method has been widely used by financial institutions to establish credit risk forecasting models.However,in the new credit business,the data with label is usually scarce,resulting in the lack of effective training data for supervised modeling.In this paper,a new transfer learning method is designed to alleviate the scarcity of training data for new cross-regional credit business,and then establish an effective credit risk prediction model.This paper improves the semi-supervised model Tri-training,and then integrates it into transfer learning,and proposes the Trans TRIT transfer model.Trans TRIT introduced confidence constraints and selected auxiliary samples according to the principle of "the minority is subordinate to the majority".Finally,ensemble learning was used to improve the generalization ability of the model.Compared to other transfer methods,Trans TRIT is more flexible to use,more robust to predict results,and superior to feature transfer model in terms of temporal and spatial performance.This paper takes the public data of Lending Club,the largest online Lending platform in the United States,as an example.The regions with a large amount of data are taken as the source domain,and the regions with a small amount of data are taken as the target domain.In this paper,six experiments were designed based on different regions and base learners to compare the prediction effect of Trans TRIT with the traditional model and other transfer models,respectively,to evaluate the effectiveness of the model.The results showed that Trans TRIT showed better performance in most regions.The model is not easily affected by the change of the base model and the data amount in the source domain.As the volume of source domain data increases,the Trans TRIT model with XGBoost is able to process mixed data more effectively,thus steadily improving the risk prediction.This paper believes that Trans TRIT model can be used to transfer credit data across regions to help financial institutions to prevent and control risks in the early stage of new credit business.