A stacked ensemble approach to identify internet of things network attacks through traffic analysis

The internet of things (IoT) has increased exponentially in connected devices worldwide in recent years. However, this rapid growth also introduces significant security challenges since many IoT devices have vulnerabilities that can be exploited for cyber-attacks. Anomaly detection using machine learning algorithms shows promise for identifying abnormal network traffic indicative of IoT attacks. This paper proposes an ensemble learning framework for anomaly detection in IoT networks. A systematic literature review analyzes recent research applying machine learning for IoT security. Subsequently, a novel stacked ensemble model is presented, combining multiple base classifiers (random forest, neural network, support vector machine (SVM)) and meta-classifiers (gradient boosting) for improved performance. The model is evaluated on the IoTID20 dataset, using network traffic features to detect anomalies across binary, multi-class, and multi-label classifications. Experimental results demonstrate that the ensemble model achieved 99.7% accuracy and F1 score for binary classification, 99.5% accuracy for multi-class, and 91.2% accuracy for multi-label classification, outperforming previous methods. The model provides an effective anomaly detection approach to identify malicious activities and mitigate IoT security threats.