Semi-Supervised Blockchain Anomaly Transaction Detection Based on Deep AutoEncoder and Multi-Layer Perceptron
Chundong Wang, Weijie Yang · 2024
Anomaly detection in blockchain transactions is crucial for ensuring the security of blockchain networks. To address the limitations of existing methods, this study proposes a semi-supervised learning model, S-DAEMLP, based on Deep Autoencoder (DAE) and Multi-Layer Perceptron (MLP), combining the strengths of unsupervised and supervised learning. The model uses a dual-structure autoencoder to independently reconstruct the structural and feature information of blockchain transaction data, further optimizing detection performance with a small amount of labeled data. Additionally, an adaptive threshold optimization algorithm is introduced to enhance model adaptability. The experimental results on the Bitcoin transaction dataset show that the S-DAEMLP model outperforms traditional methods in accuracy, precision, recall, and F1 score, demonstrating its effectiveness in detecting abnormal transactions in blockchain.