An Adaptive Framework for Real-Time Intrusion Detection in WSNs and IoMT: Comparative Study using CICDDoS-2019, WSN-DS and CICIoMT2024 Datasets

Abstract

Despite the high amount of research on networks and systems security, that comes out every day, the field of intrusion detection still attracts the interest of many researchers as one of the most important axes that contribute to understand the attacks behavior for providing an efficient and accurate solution with low cost. Wireless sensor network and Internet of Things are one of the special networks that have several constraints, limited computational resources, and the most vulnerable for attacks, especially, DoS and DDoS attacks, which considered as the most dangerous among all attacks types. Motivating by the considerations mentioned above, and using three benchmark, real-time and well-known datasets: CICDDoS-2019, WSN-DS and CICIoMT2024, we have proposed an efficient intrusion detection approach based on a lightweight learning model called LightGBM. We have compared its performance with two others machine-learning models recommended usually to intrusion detection: Decision Tree and SVM, using the metrics of evaluation: accuracy, precision, F1-score and confusion matrix. Features engineering techniques such as features extraction, features scaling and dimensionality reduction (PCA) as well as data balancing, have been used to improve the accuracy rate of our models in classifying normal and abnormal traffic. The experiment result shows that the LightGBM model is the best model among the three models for the three datasets.