MQTTNET-IDS: Deep-Fuzzy Fusion for Intelligent Threat Detection

Abstract: Message Queuing Telemetry Transport (MQTT) is a lightweight communication protocol widely adopted in Internet of Things (IoT) environments due to its low bandwidth consumption and efficient publish-subscribe architecture. Despite these advantages, MQTT-based IoT networks are highly vulnerable to cyberattacks, including denial-of-service, flooding, brute-force authentication attempts, malformed packet injection, and protocol misuse. Traditional intrusion detection systems (IDS) often rely on single machine learning models and raw traffic features, which limits their robustness and detection capability under complex and uncertain IoT traffic conditions. This paper proposes an optimized ensemble-based intrusion detection system for MQTT-based IoT networks using XGBoost and LightGBM classifiers. A multi-stage feature extraction strategy is employed, integrating raw MQTT protocol features, statistical traffic descriptors, and deep latent representations learned through an autoencoder. An optimization-driven feature selection and weighted fusion mechanism is applied to generate a compact and discriminative feature space. The final intrusion detection is performed using an ensemble of XGBoost and LightGBM to leverage their complementary learning capabilities. Experimental results demonstrate that the proposed ensemble IDS achieves high detection accuracy, low false positive rate, and strong Matthews Correlation Coefficient (MCC) across multiple test scenarios, confirming its effectiveness and reliability for securing MQTT-based IoT environments.