Improving Intrusion Detection Systems for IoT Devices using Automated Feature Generation based on ToN_IoT dataset

The Internet of Things (IoT) has witnessed exponential growth in recent years, leading to a diverse and interconnected ecosystem of devices. However, this rapid expansion has also made IoT vulnerable to various security threats and attacks. The interconnected nature of IoT devices and their extensive integration into everyday life make them enticing targets for malicious actors. Consequently, the development and deployment of effective intrusion detection systems for IoT environments have become crucial. In the literature, it has been observed that feature engineering, feature extraction, and other preprocessing steps are problematic. The general trend has been to develop intrusion detection systems using complex models such as deep learning concepts, while reducing the effort spent on feature engineering. In this study, the importance of feature engineering is addressed, and it is demonstrated that effective results can be achieved with simple models when proper preprocessing and feature generation steps are applied. An intrusion detection system for IoT devices has been implemented in the ToN_IoT dataset by employing appropriate preprocessing steps and, additionally, utilizing mechanisms for automatic feature generation. In the experiments conducted on the ToN-IoT dataset, we propose a simple model that gives comparable results with the state-of-the-art deep learning models. This model utilizes a basic random forest algorithm and benefits f rom a different t raining scheme that take the benefits of grouping, stratification, re sampling, and automated feature generation strategies. We achieved 99.99% ROC-AUC values for both train and independent test sets. The proposed method shows mostly better performances for specifity, precision, recall, and F1-score than deep learning based models.