Intrusion Detection in IoT Devices Using ML and DL Models with Fisher Score Feature Selection

IoT devices are physical things that have sensors, network connectivity, and software built into them. This allows them to gather data and share it with other systems and devices. The rising usage of Internet of Things (IoT) systems in critical infrastructure and industrial settings has raised concerns about their vulnerability to cyber-attacks. Therefore, the IIoT desperately needs techniques for enhancing strategic actions. In this work, we propose an Intrusion Detection System (IDS) for IoT devices by using deep learning (DL) and machine learning (ML) algorithms, incorporating Fisher score feature selection on the Edge-IIoT dataset. To develop effective IDS models, we have first applied the Fisher score feature selection method for the identification of the most discriminative features from the Edge-IIoT dataset. For the ML-based IDS, we employ decision tree and random forest models, optimizing their hyperparameters using a systematic hyperparameter tuning approach getting 93.7% (Decision Tree), 94.36% (Random Forest), and 94.5% (Random Forest with hyperparameter tuning) accuracy. For the DL-based IDS, we propose a single-layered feed forward neural network (FFNN) and a multi-layered feed forward neural network (MLFNN) getting 96.1% and 96.5% accuracy, respectively. These models are trained on the selected features from the dataset and evaluated using various performance indicators, including recall, F1-score, accuracy, and precision. Overall, this research work contributes to the field of intrusion detection in IoT devices by combining Fisher score feature selection with both ML and DL algorithms. The findings highlight the prospective of hybrid approaches in increasing the security and resilience of IoT systems, ultimately enabling more robust and efficient intrusion detection mechanisms for IoT deployments in critical domains.