Detect-IoT: A Comparative Analysis of Machine Learning Algorithms for Detecting Compromised IoT Devices

The rapid expansion of IoT brings unmatched convenience and connectivity, but it also raises significant security concerns. The prioritization of functionality over security in IoT devices exposes vulnerabilities like default credentials, outdated components, and insecure interfaces. To mitigate risks and combat cyberattacks effectively, it is crucial to identify and isolate compromised IoT infrastructures. In this paper, we present a curated dataset for IoT security research, which combines 40 recent IoT behavior datasets using class balancing and feature reduction techniques. This curated dataset serves as a valuable resource for future research in the field. Additionally, we compare machine learning techniques to detect compromised IoT devices, leveraging preprocessed and SMOTE-balanced network data. Our ensemble model surpasses other methods, achieving an impressive up to 98 percent F1-score, thus highlighting its efficacy in predicting compromised IoT devices and emphasizing the significance of our dataset and methodology contributions.