Implementing Machine Learning with Imbalanced Data Handling for Cyberattack Detection Using the BoT-IoT Dataset

Abstract

The rapid expansion of the Internet of Things (IoT) has significantly increased the attack surface of modern network environments, making cyberattack detection a critical component of IoT security. However, one of the major challenges in developing effective Intrusion Detection Systems (IDS) is the extreme class imbalance commonly found in real-world datasets, including the widely used BoT-IoT dataset. This study proposes a machine-learning-based framework that incorporates an imbalanced data handling technique to enhance the detection performance of minority attack classes. The methodology involves preprocessing and normalizing the BoT-IoT dataset, applying the Adaptive Synthetic Sampling (ADASYN) method to mitigate imbalance. Experimental results demonstrate that the integration of ADASYN substantially improves the sensitivity of machine learning models, particularly in detecting under-represented attacks, while also increasing overall classification accuracy and F1-score. The findings highlight the importance of advanced imbalance-handling strategies for developing robust and reliable IDS solutions in IoT environments, and provide a strong foundation for future research in large-scale, real-time IoT security systems.