Blockchain-Enhanced Anomaly Detection Algorithm for Financial Data Sharing Platforms Using u-BlockMixup

A Smart Contract–Driven Machine Learning Framework for Secure and Reliable Financial Data Sharing

Abstract

As the financial industry advances in digitalization, the security and quality of shared data have become critical. Traditional centralized anomaly detection methods face challenges regarding single points of failure and data tampering. To address these issues, this paper proposes a Blockchain-Enhanced Anomaly Detection Algorithm using u-BlockMixup. We integrate a decentralized blockchain architecture with a semi-supervised deep learning model based on the Mean Teacher framework. Specifically, we introduce the u-BlockMixup data augmentation method, which combines supervised cross-entropy loss and unsupervised consistency loss to generate high-quality synthetic samples, thereby improving generalization on limited labeled data. Experimental results on a dataset of over 1 million financial transaction records demonstrate that the proposed method outperforms traditional Isolation Forest and Autoencoder models. The algorithm achieves an accuracy improvement of 15% and an F1 score increase of 18%, with a false positive rate reduced to 2.3%. These findings confirm that combining blockchain immutability with u-BlockMixup-enhanced machine learning significantly improves the reliability and real-time detection capabilities of financial data sharing platforms.