Transaction fraud detection algorithm based on graph neural network and user behavior

With the diversity of transaction forms continue to rise, and transaction fraud is increasingly widespread. The traditional way to deal with transaction fraud depends on the rule system and manual verification, which is not only inefficient, but also difficult to cope with complex and changeable fraud routines.This paper proposes an improved GraphSAGE network based on multilateral types, integrating the attention mechanism of multilateral types and high-order graph convolution. The multi-head edge type attention mechanism can dynamically adjust the aggregate weight of neighbor node information according to the edge type, so as to capture the differential contribution of different edge types to the node representation. The higher-order graph convolution operation can capture more distant dependencies in the graph. In order to enhance the interpretability of the model, this paper also introduces the SHAP value analysis method to explain the basis of the model prediction results. Experimental data show that the improved GraphSAGE network has excellent performance and higher AUC than the traditional method.The research results of this paper are expected to be promoted and applied in more fields, such as the access scenario of aggregation chain.