Guarding Against IoT Threats: An Analysis of Intrusion Detection with the Kitsune Attack Dataset
Anshika Sharma, Himanshi Babbar · 2024
The extensive adoption of Internet of Things (IoT) devices has resulted in previously unheard-of levels of connectedness and ease, but it has also created new cybersecurity challenges. Hackers are turning an increasing number of IoT devices into their targets. They take advantage of security flaws to execute a variety of attacks, from botnet malware attacks and distributed denial-of-service (DDoS) attacks to surveillance, reconnaissance (recon.) and man-in-the-middle(MITM) attacks. Innovative detection techniques are required because traditional security systems typically cannot keep up with the dynamic and varied nature of IoT environments. Using the Kitsune attack dataset, a large collection of network traffic captures that have been carefully chosen for use in network intrusion detection system (NIDS) research, the effectiveness of machine learning (ML)-based techniques have been examined for identifying IoT attacks in this study. Utilising characteristics taken from network traffic data, such as communication patterns, payload attributes, and packet headers, investigate how well-supervised learning algorithms like Support Vector Machines (SVM), Random Forest (RF), Logistic Regression (LR), and K-Nearest Neighbours (KNN) perform. Each algorithm's detection performance has been assessed using metrics including accuracy, precision, recall, and F1-score. Their respective advantages and disadvantages in terms of scalability, computational efficiency, and detection accuracy have also been compared. Additionally, the effects of feature selection and ensemble learning approaches have been evaluated on detection performance, offering recommendations for constructing durable and dependable IoT-IDS. The results show that, at 98.99%, the LR model has the highest accuracy. On the other side, the accuracy rates of the KNN, SVM and RF models are 79.42%, 92.75%, and 97.98%, respectively.