Increasingly frequent illegal transactions have seriously infringed on the security of Ethereum transactions, while the anonymity of e-currencies makes it difficult to track and analyse illegal transactions, and the detection of anomalous accounts is crucial for maintaining its security and stability. In this paper, we propose a multi-node distributed framework based on the integrated algorithm of XGBoost with ADASYN for efficiently and accurately detecting anomalous transactions in Ethereum. Finally, the experimental results show that the F1-score value of the XGBoost with ADASYN integrated algorithm based on distributed training proposed in the article is 99.63%, which is more efficient and accurate compared with single-node XGBoost, CatBoost and other methods, and effectively improves the performance of Ethereum anomaly detection.