Suspicious transaction identification is an important element of anti-money laundering work, and it has become a new trend to use algorithmic models as tools to analyze and identify suspicious transactions. Deep convolutional neural networks can effectively and automatically extract classification features from data, and have been widely used in various fields of research as they have shown good recognition results in many classification tasks. In this paper, we firstly design an 8-layer model framework based on Alexnet network theory for suspicious transaction identification analysis. Secondly, the Elliptic data set is divided into training and testing sets in the ratio of 7:3, and the divided data are used to train and test the model. Finally, the robustness of the model to data input is explored by randomly disrupting the ordering of the elements in the input data. The results show that the Alexnet convolutional neural network has good applicability for suspicious transaction recognition, and the overall classification accuracy of the Elliptic dataset can reach 94%, and the proposed model in this paper has good recognition accuracy in general.

Alexnet, part defects, deep learning, board parts