A Bitcoin Illegal Transaction Detection Method Based on Graph Neural Networks

The anonymity and decentralization of Bitcoin make it a significant medium for illicit transactions, posing challenges for traditional detection methods in handling complex transaction network structures. This study proposes a graph neural network model based on a pre-trained Conditional Variational Autoencoder (CVAE) to enhance the efficiency and accuracy of Bitcoin illicit transaction detection. The model generates K−1 feature vectors through the CVAE, which have the same number as the input features, and then combines these generated K−1 feature vectors with the original feature vector to ultimately form K feature vectors. Each feature vector undergoes multi-channel aggregation and max pooling, resulting in multiple feature vectors. These vectors are subsequently processed through linear layers and layer normalization, followed by another round of max pooling to obtain a global feature vector. Finally, the feature vectors are further processed through graph convolutional layers and linear layers to generate the final classification result. The model integrates input feature vectors at the output layer through a skip mechanism. Experimental results demonstrate that this model performs excellently in Bitcoin illicit transaction detection, significantly improving detection accuracy and robustness.