Federal learning method and device for credit card fraud prevention

Abstract

The invention discloses a federated learning method and device for credit card fraud prevention. The method comprises the following steps: constructing local graph convolutional neural network models corresponding to K federated learning participants with different fraud categories; performing federated learning training by using the local graph convolutional neural network models, wherein the aggregation process of federal learning parameters is improved by adopting an attention mechanism, so that each local graph convolutional neural network model has a weight matched with the local graph convolutional neural network model for aggregation; and outputting a global graph convolutional neural network model, the global graph convolutional neural network model being used for processing imported user data and identifying corresponding fraud categories. Aiming at the existing problems of an existing credit card fraud evaluation method and a classical federated learning algorithm, the invention provides a federated learning algorithm suitable for non-Euclidean space data and participant individuation characteristics to process financial data and carry out credit card anti-fraud judgment.

Description

technical field [0001] The present invention relates to the technical field of credit card anti-fraud, in particular to a federated learning method and device for credit card anti-fraud. Background technique [0002] With the rapid development of the Internet, financial technology based on artificial intelligence technology has a profound impact on people's consumption behavior. However, financial data often involves privacy, and the data of different banks, lending institutions and other financial institutions cannot be directly shared, which forms data silos. However, in order to achieve high accuracy of artificial intelligence algorithms, a large amount of data is required, and the algorithm model independently trained by a single data owner cannot accurately assess whether credit card fraud is present. [0003] As a distributed machine learning / deep learning framework that protects data privacy, federated learning can provide a good solution to data silos, serious data ...

AI Technical Summary

Problems solved by technology

The existing federated learning algorithm only averages the parameters of each local model. First, it does not take into account the personalization of each local model (for example, for an anti-fraud analysis system, in the sample, users of different financial institutions The characteristics of the data are inconsistent, and the overall amount of the loan is also different according to the level of the regional economy), which cannot cope with the influence of different sample centers of characteristics due to the different environments of each client; the second is that it does not take into account the data in the actual environment Most of them are in non-Euclidean space, such as the relationship between users, financial knowledge graph, etc. Because the evaluation standards of non-Euclidean space data are inconsistent and the data structure is irregular, it is difficult to obtain high-performance models for joint training of such data

Such problems lead to the shortcomings of not being accurate enough in learning

Images