In recent years, blockchain has rapidly evolved, becoming a crucial foundation for cryptocurrencies and digital assets. However, this growth has brought new risks like money laundering and fraud. Ensuring user protection and detecting suspicious accounts is now a vital concern in blockchain development. Therefore, detecting abnormal transactions in blockchain has emerged as an important research field. This study proposes a machine learning-based approach to identify fraudulent activities in phishing accounts. Real data from Ethereum is utilized, specifically selecting transaction data from accounts marked as phishing accounts by Etherscan, as well as other unmarked accounts for comparison. Machine learning techniques are employed to assist in identifying abnormal transaction behaviors, with the ultimate goal of enhancing the security of blockchain transactions. The thesis introduces transaction time and frequency as a characteristic for identifying phishing accounts, and the verification results validate this idea.