Hybrid Feature Selection Based Lightweight Network Intrusion Detection System for MQTT Protocol

The proliferation of Message Queuing Telemetry Transport (MQTT) communication on the Internet of Things (IoT) has raised significant security challenges, a lightweight Network Intrusion detection System (NIDS) model is needed to reduce the complexity of the process in IoT modules. This paper proposed a NIDS based hybrid feature selection, combining XGBoost and MaxPoolingID. The proposed algorithm enhances the MQTT-enabled IoT security by effectively capture the complex pattern and extracts relevant features. The algorithm is experiment in two scenarios which is the uniflow and biflow of the MQTT dataset. The results are shown using 4x4 confusion matrix. In both uniflow and biflow scenarios, the model achieves outstanding accuracy, precision, recall and F1-Score, exceeding 90%. The biflow scenario outperforms the uniflow across all matrixes, making it preferable for network data classification.