Research on Blockchain Anomalous Behavior Detection Technology

For blockchain transactions, how to distinguish normal transactions from abnormal transactions is a very critical issue. To this end, we introduce deep learning as a capable tool for abnormal behavior detection detection, and its excellent learning ability shines in this field. In this paper, we propose a deep learning time series Transformer model based on deep learning combined with ResNet (Residual Network) for anomaly detection and classification of Bitcoin dataset. The Elliptic dataset is used and the dataset is top-filled by manually crawling the latest Bitcoin dataset, and the Random Forest algorithm is used for feature selection, and finally 16 features are selected for classification. After testing under different sampling methods the Transformer model combined with residual network shows good detection results compared to structures such as single machine learning algorithm SVM (Support Vector Machine) and other deep learning models CNN (Convolutional Neural Network).