Enhancing User Anonymity in IoT Blockchain: Machine Learning Deanonymization Risks and Mitigations
Sapan Bharadwaj Bonala, Sharath Chandra Macha, Sukender Reddy Mallreddy, Yeshwanth Vasa · 2024
In the context of Internet of Things (IoT) and smart grids—this research studies the anonymity of blockchain users— with a particular emphasis on Machine Learning (ML) techniques for user de-anonymization. Blockchain data is eternal—yet user privacy is still sensitive. While concerns about anonymity are addressed by existing research—more study is required for IoT and smart grids. Therefore, this study highlights long-term anonymity threats owing to permanent data by analyzing blockchain transactions and off-chain sun exposure data to evaluate ML's efficacy in identifying smart grid blockchain users. The process includes gathering data from the power grid, creating blockchain ledgers, and using ML on historical transactions to identify households. In order to improve privacy—it assesses the value of including off-chain solar data and evaluates ledger obfuscation and public key techniques. As demonstrated by the data, ML models are able to reliably link user transactions. Using a convolutional network—peak classification accuracy was 84% for customer ID and 63% for postcode. Attacker success rates are increased when solar exposure data is included. Household energy consumption and solar generation are 85% accurately reconstructed by regression models. Public keys drastically lower attack success—with customer ID and postcode having the lowest classification accuracy drops to 4% and 11%, respectively.