Adaptive AI-Blockchain Framework for Threat Detection and Mitigation in IoT-Enabled Smart Cities

Abstract

The growth of Internet of Things (IoT) devices in smart cities has increased the risk of cyber threats due to new attack surfaces and limited built-in security. This paper presents an adaptive, multi-layered security framework that combines deep learning, reinforcement learning, and blockchain to detect and mitigate threats in IoT environments. The system uses a Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) model to detect anomalies in network traffic with high accuracy. When a threat is detected, a Deep Q-Network (DQN) agent selects the most appropriate response based on the threat level, trust score of the device, and service importance. To ensure transparency and integrity, all events and actions are stored immutably using Hyperledger Fabric and the InterPlanetary File System (IPFS). Experiments using the BoT-IoT dataset show that the CNN–LSTM model achieves 97.8% accuracy and an AUC of 0.992. The DQN agent reduces false isolations to 2.8% and maintains an average response time of 148 ms. Compared to traditional systems, the proposed framework offers better accuracy, faster decision-making, and improved trust management. The use of blockchain ensures secure, auditable records across multiple domains. This approach provides a scalable and intelligent solution for securing smart city infrastructures against evolving threats.