DEVELOPMENT OF AN IOT-BASED INTRUSION DETECTION SYSTEM USING MACHINE LEARNING

ABSTRACT The Internet of Things (IoT) has emerged as a transformative technological paradigm, connecting billions of devices across sectors such as healthcare, manufacturing, and smart cities. However, this rapid expansion introduced critical security vulnerabilities, as IoT devices often operated with limited resources, weak security protocols, and outdated firmware. Consequently, IoT environments were increasingly targeted by cyberattacks such as Distributed Denial of Service (DDoS), botnet infiltration, and unauthorized access, which traditional security tools failed to detect or mitigate effectively. Existing Intrusion Detection Systems (IDS) exhibited limitations in accuracy and scalability when applied to IoT-specific traffic, leaving networks exposed to undetected threats. To address these challenges, this study developed a machine learning–based IDS framework using a Random Forest classifier to enhance threat detection and improve IoT resilience. The methodology adopted an experimental research design and leveraged the UNSW-NB15 dataset, containing realistic benign and malicious IoT network traffic samples. Data preprocessing involved removing duplicates, encoding categorical variables, and normalizing features for consistency. The Random Forest model was trained on labeled binary data (0 = Normal, 1 = Malicious) and deployed through a Streamlit web application, which provided an interactive interface for dataset upload, preprocessing feedback, and visualization of detection results. The model’s performance was evaluated using standard metrics, achieving an accuracy of 99.74%, a precision of 99.74%, and a recall of 99.88%. Analysis of the confusion matrix and prediction distribution demonstrated that the system successfully identified nearly all malicious traffic, with only a minimal number of false negatives. This highlights the robustness and reliability of the proposed framework. The IoT SecureDetect system provides a practical, interpretable, and reproducible solution for IoT intrusion detection, while the Streamlit interface enhances usability for both technical and non-technical users. Overall, this study contributes a highly effective methodology for AI-driven IoT security and establishes a foundation for future enhancements, including hybrid modeling, dataset expansion, and real-time deployment.