Credit card transaction risk prediction method based on federated learning

Abstract

The invention relates to a federated learning-based credit card transaction risk prediction method, and the method comprises the following steps: 1) obtaining a data set about client credit card transaction feature data in each bank serving as different local clients; 2) searching similar instances in each bank data set by adopting a locality sensitive hash algorithm; 3) constructing combined features by adopting a GBDT algorithm in a serial federated learning environment; 4) constructing a new training feature according to the combined feature and the original feature, and expanding and constructing a new data set by each local client; 5) enabling each local client to adopt the same neural network model for training, uploading the trained model parameters to the cloud, and enabling the cloud to aggregate and update the model parameters and returns the model parameters to each local client to start the next training until the training process converges, so that the final neural networkmodel is obtained to complete a credit card transaction risk prediction result. Compared with the prior art, the method has the advantages of privacy protection, accurate result, wide application range and the like.

Description

technical field [0001] The invention relates to the field of financial data security, in particular to a credit card transaction risk prediction method based on federated learning. Background technique [0002] In recent years, federated learning has been flourishing as an emerging basic technology of artificial intelligence. The concept of "federated learning" was first proposed in 2016 by Google research scientist H. Brendan McMahan and others. It refers to the setting in which multiple clients (such as mobile devices, institutions, organizations, etc.) collaborate with one or more central servers to perform decentralized machine learning. In the process of decentralized machine learning, federated learning can ensure that the private data of each customer is not stored locally, thereby reducing the risk of privacy leakage and the corresponding cost caused by traditional centralized machine learning. In addition, according to the bank's annual reports in recent years, th...

AI Technical Summary

Problems solved by technology

[0006] First of all, the improvement brought about by these works is relatively limited, because the data used are all from a single institution, and the data barriers between institutions have not been broken, so it is impossible to achieve the effect of inter-agency cooperation and win-win;

[0007] Secondly, in order to protect data privacy, some works apply the differential privacy method, introduce noise in the training process, and may add some unstable modules, making it difficult to train the entire model or the training time is too long, reducing the training accuracy , not applicable to actual scenarios;

[0008] Finally, some works only use existing single features, while ignoring the importance of combined features, which has a certain impact on the training results of the model

Images