Using Hybrid Neural Networks to Detect Tax-Evading Blockchain Transactions Disguised via Encrypted DNS Payloads

Abstract

The pseudonymous nature of blockchain transactions, combined with the rise of encrypted DNS protocols such as DNS-over-HTTPS (DoH) and DNS-over-TLS (DoT), has created a new frontier for sophisticated tax evasion. Malicious actors can now exfiltrate transaction details and coordinate transfers by encoding data within the payloads of encrypted DNS queries, effectively bypassing traditional network monitoring and forensic analysis. This paper proposes a novel detection framework that leverages a hybrid deep learning architecture to identify such covert, tax-evading activities. Our system integrates a Convolutional Neural Network (CNN) for its superior ability to extract spatial and sequential patterns from raw network flow data and encrypted payload characteristics, with a Long Short-Term Memory (LSTM) network to model the temporal dynamics of blockchain interactions and DNS query sequences. By fusing these two paradigms, the hybrid model can distinguish between benign encrypted DNS traffic and malicious payloads used for illicit financial coordination. We evaluate our framework using a synthetically generated dataset that simulates realistic tax-evasion strategies, including micro-transaction splitting and delayed transaction relaying. Preliminary results indicate that our approach achieves a significantly higher detection rate and lower false-positive rate compared to conventional signature-based or single-model machine learning methods. This research demonstrates the efficacy of hybrid neural networks in preserving financial integrity and provides a critical tool for regulatory agencies to enforce tax compliance in the age of encrypted communications and decentralized finance.