<p>The development of IoT devices of various kinds has caused the amount of data generated to be explosive, raising several challenging issues, such as dynamic attack patterns for threat prediction, data privacy issues, and constrained computing resources at the network edge. Central IDS and Static machine learning models do not scale, adapt to new threats, or maintain the confidentiality and integrity of the data. However, most federated learning mechanisms lack trust in model aggregation and fail to prioritise interpretability. As such, a real-time, secure and explainable IoT threat analytics is a research need. To alleviate these limitations, this work proposes ThreatFedChainAI, an architecture dedicated to threat anticipation and detection for IoT that integrates an edge-blockchain. The proposed model employs a quantum-inspired particle swarm optimisation algorithm to solve the feature selection problem and filters out irrelevant features from edge nodes by reducing data dimensionality with minimal loss of robbery-related information. An adaptive federated learning method is combined with blockchain smart contract validation to ensure secure, tamper-proof, and privacy-preserving model updating. In addition, SHAP-based interpretability techniques are utilised to increase the explainability of model predictions. Experimental results on the CICIDS2017 and TON_IoT datasets show that our ThreatFedChainAI outperforms baseline models by up to 5.3% in accuracy and that F1-scores are consistently above 97%. The effectiveness of the proposed system is validated through ablation studies, and visualisations and tables are presented for interpretation. Overall, the proposed system introduces a scalable, secure, and interpretable approach for real-time IoT threat detection that addresses first-rank issues in privacy, trust, and adaptability. This makes ThreatFedChainAI ideally suited for deploying at scale in mission-critical IoT networks.</p>