Towards optimized machine-learning-driven intrusion detection for Internet of Things applications

As the utilization of Internet of Things devices becomes increasingly important in real-life applications, the need to address potential threats to the infrastructure of these networks, such as security breaches and intrusions, has also grown. While intelligent systems for detecting attacks are often considered solutions to combat these threats, many of these systems lack the accuracy to effectively respond to evolving versions of these attacks and lack flexibility. In addition, constructing reliable intelligent systems for detecting intrusions relies heavily on the availability of valid datasets. In this study, an optimized machine learning-based model for intrusion detection in the IoT field is proposed. The model utilizes a collection of high-quality real-time data to train and test the performance of the proposed machine learning. based model, and the results are compared among a selection of ML algorithms. The algorithms used in the study include Fine and Optimizable Tree, Optimizable Discriminant, Optimizable KNN, Logistic Regression Kernel, and Optimizable Ensemble. The experiments were conducted on the NLS-KDD dataset using the MATLAB platform to evaluate the effectiveness of the proposed model, focusing on performance metrics, including accuracy, F1-score, precision, recall, and confusion matrix. A comparative analysis is performed by comparing the results with existing research findings. The findings highlight that the proposed model using the Optimizable Ensemble algorithm achieves best performance in terms of accuracy (99.9%).