IJAEMS
IJAEMS
The rapid deployment of machine learning models on edge devices has enabled real-time intelligent applications in domains such as healthcare, smart transportation, industrial automation, and Internet of Things (IoT) systems. The use of machine learning models on edge devices has opened up the door to intelligent applications in real time across several areas, including healthcare, smart transportation, the Internet of Things (IoT), and industrial automation. The distributed and resourceconstrained nature of edge environments, however, makes machine learning pipelines susceptible to adversarial attacks that can alter input data and lead to wrong predictions. In this paper, a framework for real-time adversarial attack detection in edge machine learning pipelines using Explainable Artificial Intelligence (XAI) is proposed. The proposed approach extends the monitoring of incoming data and the behavior of the model, looking for unusual patterns linked to adversarial perturbations, and, additionally, adds an explainability layer, providing interpretable information about what features and decision factors are behind the detection outcome. It combines lightweight machine learning, adversarial sample analysis, and real-time detection and explanation with SHAP to provide low-latency security monitoring on edge devices with limited resources. The results from experimental testing with the Edge-IIoT-set dataset are shown to be effective in distinguishing between normal and malicious network traffic with an overall classification accuracy of 97.71%. The proposed solution offers an interpretable and reliable security mechanism to secure edge machine learning pipelines against adversarial attacks while preserving the requirements of real-time edge intelligence.