Improving the IDS for BoT-IoT Dataset-Based Machine Learning Classifiers

In this paper, we use four supervised Machine Learning (ML) classifiers, including Naive Base (NB), Random Forest (RF), Gradient Boost (GB), and Decision Tree, to analyze the current BoT-IoT dataset. Instead of considering outdated datasets such as the NSL-KDD or others, we based our study on a new dataset dedicated to IoT networks. This allows us to analyze the cybersecurity challenges in IoT networks using ML algorithms. The Bot-IoT dataset is known as an imbalanced dataset where some classes of the data suffer from the lack of samples, which makes the design and the training of an ML model on this dataset a challenging task. Despite these challenges, we successfully implemented the above-mentioned ML models to analyze the cybersecurity attacks in IoT networks and show the importance of ML models trained on the Bot-IoT dataset to improve the detection rate of intrusion detection systems (IDS). Moreover, our study covers both binary and multi-classes classification for complete analyses of the IDS system.