A Survey of Deep Learning-Based Methods for Detecting Anomalous Transactions in Blockchain

Abstract

The decentralized and anonymous nature of blockchain technology has driven the growth of the digital economy while introducing new challenges in detecting anomalous transactions. This paper systematically reviews recent advances in deep learning-based methods for identifying anomalous transactions in blockchain. It analyzes the main types of blockchain-based anomalous transactions and their risks, highlighting the limitations of traditional detection methods in handling high-dimensional, nonlinear transaction data. Furthermore, this survey provides a detailed introduction to six major categories of deep learning approaches: graph neural network-based methods, which effectively model transaction network topologies; autoencoder-based methods that identify anomalies through reconstruction errors; generative adversarial networks that mitigate data imbalance issues; attention mechanisms capable of capturing critical transaction features; temporal models suited for analyzing transaction timing patterns; and multi-feature fusion methods that enhance detection comprehensiveness. For each type of method, this paper summarizes the representative research achievements and their performance，and discusses the key challenges faced by current research and proposes future research directions.