Research on Blockchain Security Detection Based on Machine Learning

With the rapid development of blockchain technology, its decentralized, tamper-resistant and traceable characteristics have been widely adopted in fields such as finance, supply chain management and the Internet of Things. However, the openness of blockchain ecosystems also creates opportunities for criminals, posing serious threats to the stability of blockchain networks and the security of user assets. To address the difficulties in identifying on-chain malicious transactions, the low audit efficiency of smart contract vulnerabilities, and the complexity and poor reproducibility of existing models in traditional blockchain scenarios, this paper proposes a lightweight machine learning-based detection scheme. The Elliptic transaction dataset and a smart contract vulnerability dataset constructed from the Slither official test suite and SWC (Smart Contract Weakness Classification) standard vulnerability cases are adopted, and random forest, XGBoost, LSTM and CNN are selected to build detection models. Multi-model comparative experiments are conducted with accuracy, precision, recall, F1-score and AUC as evaluation metrics. Experimental results show that the proposed scheme can effectively detect malicious transactions and contract vulnerabilities, providing a reference for intelligent blockchain security protection.