Trustworthy Anomaly Detection in Industrial IoT-Enabled Cyber-Physical Systems Using Explainable Ensemble Learning

Abstract

Cyber-physical manufacturing systems based on industrial IoT generate large amounts of diverse sensor data. This makes them more vulnerable to operational abnormalities that can lead to equipment failure, safety risks, and produc-tion downtime. Intelligent manufacturing systems must therefore have accurate and reliable anomaly detection. This paper suggests a credible system of anomaly detection with the use of explainable ensemble learning on cyber-physical systems of industrial IoT. The given solution combines mul-tiple machine learning models, such as XGBoost, logistic re-gression, random forest, support vector machine, and multi-layer perceptron, into the stacking ensemble architecture to consider the different characteristics of data and minimize the bias of individual models. The probabilistic outputs of the base learners are combined using a meta-model, which leads to better detection accuracy and high strength. Through experiments applied to a real-world smart manufacturing IoT dataset, it can be observed that the proposed ensemble model has an accuracy of 97.98%, which is the highest of all the single classifiers and balanced across normal and anomalous classes. Moreover, the explainable AI with SHAP is imple-mented to allow transparency into the decision-making pro-cess. The findings validate that the proposed framework pro-vides a scalable, reliable, and interpretable solution to real-time anomaly detection in industrial IoT-enabled cyber-physical manufacturing systems.