Blockchain technology has attracted significant attention in recent years for its potential to revolutionize various industries. However, its decentralized and pseudonymous nature also exposes it to fraudulent activities, highlighting the need for effective anomaly detection techniques. In this paper, we propose a novel blockchain transaction anomaly detection algorithm called H-Regularization with Mean Normalization (HRM), which is based on a one-class learning framework. Specifically, it utilizes a three-layer Multilayer Perceptron (MLP) model which combines holistic regularization mechanism with mean normalization to detect abnormal transactions within the blockchain network. Experimental results on the real-world datasets demonstrate that HRM outperforms traditional anomaly detection methods, including Logistic Regression and Random Forest, in terms of recall and F1 score. Our method’s simplicity and effectiveness make it a promising solution for enhancing security and transparency in blockchain transactions.