Blockchain-Enabled Fault-Tolerant IoT Intrusion Detection Using Spatio-Temporal Deep Learning

Abstract

Due to the swift expansion of the IoT networks, there are some critical security issues such as node compromising, data manipulation, and Sybil attacks. This paper presents a fault-tolerant IoT security architecture based on blockchain and multi-layer anomaly detection with secure consensus-based systems. The IoT devices are authenticated via a blockchain-based identity verification and through cryptographic authentication to verify legitimate participation, sensor data is encrypted, timestamped and logged into blockchain to prevent tampering. Data cleaning, normalization, and outlier elimination are also used as preprocesses to the ToN-IoT dataset in order to improve the quality of the data. Temporal and statistical characteristics are derived to identify time series trends as well as cross-node associations and a new Spatio-Temporal Reinforced Swarm Optimization (ST-RSO) algorithm is used to select the most pertinent features, lowers the dimension, and enhances detection rates. The detection of anomalies is carried out with the help of a Multi-Scale Residual Graph Attention Network (MS-RGAN), which captures both local and global interactions among devices with the flow of time. Verification Alerts are validated with a Majority Voting Ensemble (MVE) to minimize the number of false positives and verified events are stored on the blockchain with PBFT consensus, which ensures decentralized and resistant to tampering decision-making. Nodes that are compromised are automatically isolated, updated on trust scores and reintegrated after verification. The proposed MS-RGAN model has a high level of performance with a high accuracy of 98.7, a high precision of 98.4, high recall of 98.6, and F1-score of 98.5 and a very low FPR of 0.9 and FNR of 1.2. It has a high MCC of 96, 98.2% NPV, lower detection latency (21 ms), 41.8% feature drop, and very low consensus validation time (14.2 ms) hence it is highly efficient and reliable in intrusion detection on the IoT. The review shows that the framework can be used to achieve high detection, low false alarms, and robust performance in the presence of distributed IoT threats.