Secure Multi-Tier Deep Learning Model with Blockchain Consensus for Intelligent IoT Traffic Anomaly Detection

The exponential expansion of IoT ecosystems has led to unprecedented network traffic volumes and increased susceptibility to cyberattacks, including botnet propagation, DDoS flooding, man-in-the-middle intrusions, and device-level exploitation. Centralized intrusion detection mechanisms have proven inadequate due to limited scalability, single-point failure risks, and the lack of data privacy guarantees. To address these challenges, the proposed research introduces a Secure Multi-Tier Deep Learning Model integrated with Blockchain Consensus for intelligent anomaly detection in IoT traffic data. The architecture employs a three-tier deep learning pipeline consisting of (i) Traffic Feature Extraction using 1D-CNN, (ii) Temporal Behavioral Modeling using Bi-LSTM, and (iii) Attack Classification using Dense Multi-Head Attention, enabling high-resolution identification of complex cyber threat patterns. For security and trust, model updates and anomaly detection logs are validated and stored through a Practical Byzantine Fault Tolerance (PBFT) blockchain consensus, eliminating model-poisoning attempts and ensuring tamper-resistant collaboration across distributed IoT nodes. The proposed model is evaluated using the IoT Traffic dataset which contains multi-class attack and normal traffic samples, including Mirai, DDoS UDP/TCP, Port Scanning, DNS poisoning, Backdoor, and OS fingerprinting traffic patterns. The proposed architecture achieves 98.72% accuracy, 97.94% F1-score, 98.11% recall, and a 31.5% reduction in false positives, while decreasing communication overhead by 27.8% due to selective multi-tier update compression. Blockchain validation adds an average consensus latency of only 0.83 seconds, enabling real-time deployment in resource-constrained IoT environments.