Enhancing Machine Learning-based Anomaly Detection for IoT Networks
AbedlRahman Almodawar, Ashraf Ahmad · 2025
Anomaly detection is essential for protecting against malicious activities in the Internet of Things (IoT) settings. Fortunately, Machine Learning (ML) provides powerful mechanisms to identify deviations from normal network behavior, which can signal several security threats. However, a significant challenge is the prevalence of imbalanced traffic, where benign traffic often outweighs malicious traffic, representing the minority class, which may result in poor performance. Additionally, metrics like accuracy can be misleading in imbalanced scenarios, as a high accuracy might reflect the correct classification of the majority while failing to detect the class of interest. Therefore, to effectively train ML models for detecting anomalies in IoT networks, this research proposes handling data imbalance using oversampling techniques combined with learning optimization using hyperparameter tuning, aiming at achieving high F1 scores as well as accuracy. Using the IoT-23 benchmark dataset, our evaluation results demonstrated high and consistent scores across different metrics, compared to others.