Enhanced Intrusion Detection System in IOT-MQTT Using Machine Learning Classifiers

This study significantly enhances IoT security by employing state-of-the-art techniques to detect a range of attacks, including as DoS, brute force, and malformed packets. The hybrid classifier that combines a 1D CNN and fuzzy logic is optimized via the Improved Vulture Starvation-based African Vultures Optimisation Algorithm (IVS-AVOA). The system further enhances attack detection by employing an ensemble classifier that uses models such as Random Forest, XGBoost, and LightGBM, augmented by TSO. The hybrid solution has a 98.5% detection accuracy and a 1.2% false-positive rate, which is better than standalone rule-based orML-driven IDS. The solution offers a robust and adaptable defence mechanism, improving security in vast IoT contexts by utilizing a publically available MQTT dataset and scalable, efficient, and dependable methodologies. Keywords: IoT Security, MQTT protocol, IDS, Machine learning classifiers (Fuzzy Logic, 1DCNN)