Stacking-Based Machine Learning Approach for Anomaly Detection in Embedded System Network Traffic

Embedded systems are increasingly deployed in critical infrastructures, making them prime targets for cyber threats and malicious activities. Traditional embedded security techniques often fail to protect against sophisticated network-based attacks in real-time. In this paper, a stacking-based machine learning approach for identifying network traffic anomalies in embedded devices is proposed. It uses a dataset that includes regular and unusual network traffic patterns. For better detection, the proposed model uses stacking ensemble learning with several base classifiers. Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Stochastic Gradient Descent (SGD) are chosen as basic learners. These approaches are chosen as they can discriminate between regular and harmful network activity better. After integrating these basic learners, a meta-classifier refines the final predictions by learning from their collective outputs. With 94.4% accuracy, the stacking ensemble technique detects network abnormalities better than individual models. In contrast to solo classifiers, stacking decreases false positives and enhances anomaly identification. In embedded systems, this work delivers scalable and effective security threat detection. A stacking-based architecture can be utilized for industrial control, IoT, and vehicle security.