Secure Transactions in IoT Network Using PINN-Based Intrusion Detection System and HDPoA Blockchain Protocol

Abstract

IoT is a rapidly developing technology with a wealth of creative application possibilities. However, IoT wireless sensor networks are vulnerable to Denial of Service (DoS) attacks due to their insecure nature. Although network integrity and security have been ensured through the use of distributed ledger and blockchain technologies, privacy preservation concerns frequently arise with traditional approaches. So, deep learning-based Physics-informed neural networks (PINN) and Honesty-based Distributed Proof-of-Authority (HDPoA) are developed to enhance transaction security and detect intrusions. Initially, the mobile nodes were deployed in different regions to gather transactions and an intrusion detection system to analyze attacks. First, the Intrusion Detection System (IDS) uses a deep learning approach for detecting the intrusion in the network. For that, the collected data from the deployed nodes are pre-processed using Variational auto-encoder and min-max normalization to standardize input dataset values. Then the features are selected using wild horse optimization and classified using PINN to predict data attack or non-attack. After that, Homomorphic variable tag generation is used for normal transactions with multiple copies in the same document, which are then converted into hash values using the Keccak hashing function. The miner validates transactions based on rank-based priority. Honesty-based Distributed Proof-of-Authority (HDPoA) was used for network security, making it suitable for deployment in blockchain-based IoT applications. The proposed deep learning-based PINN classifier reached 97.2% accuracy and 96.52% specificity. Homomorphic variable (HV) tag generation takes 0.4 s, while the Keccak algorithm takes 0.3 s for hash generation, and the HDPoA protocol has 420 s for block generation time.