Intrusion Detection in IoT Environment Using Hyperparameters Tuned Machine and Deep Learning Models on the CICIoT2023 Dataset

Abstract

Internet of Things (IoT) technology has made our life connected, simple and smart by integrating physical objects to the internet in various fields. These are also systems capable of creating and transmitting data to users through various services. Since these objects with their limited resources are interconnected with each other via the internet, they are then vulnerables to many attacks. Given the constraints already mentioned, traditional intrusion detection systems (IDS) are inadequate and no longer sufficient. In this paper, we propose an intrusion detection taking on consideration the limited resources of IoT devices, using machine learning and deep learning combined with features engineering, data balancing method and hyperparameters tuning to achieve the best result. Using a wide range of evaluation metrics, including Accuracy, Precision, Recall, F1-score, confusion matrix and execution time, we have evaluated various machine and deep learning models including Support Vector Machine, Random Forest, VGGNet and Deep Neural Network, as well as an approach for features extraction such as features scaling and transformation. This study is carried out using a well-known, benchmark and real time dataset CICIoT2023, generated by IoT devices that includes thirty-three attacks, classified into seven categories, namely DDoS, DoS, Recon, Web-based, Brute Force, Spoofing, and Mirai.The experiment result demonstrates the effectiveness of Random Forest that accomplished with 91,89% in the accuracy and 92% in the precision, outperformed the others modeles in classifying attacks from normal traffic with a minimum time of execution.