Edge-AI-Based Real-Time Anomaly Detection for Industrial IoT Networks

Abstract

This study investigated the effectiveness of edge-based artificial intelligence for detecting abnormal behavior in Industrial Internet of Things (IIoT) networks. Network traffic datasets containing variables such as packet rate, protocol type, connection duration, source and destination behavior, and traffic volume were used to train and evaluate three machine-learning algorithms, Random Forest, XGBoost, and Support Vector Machine, for anomaly classification. To see whether edge processing could sharpen real-time anomaly detection without relying as heavily on centralized servers, we pitted edge and centralized deployments against each other, measuring accuracy, precision, recall, F1-score, latency, and resource use. All three models classified traffic well across both deployment modes, though XGBoost turned in the best overall detection accuracy and F1-score. Moving processing to the edge came with a small drop in detection accuracy, but the steep drop in latency and network bandwidth made that trade-off worthwhile for real-time responsiveness. These outcomes show that edge-based AI is a practical, advantageous route for real-time anomaly detection when Industrial IoT networks cannot tolerate delay, and they offer clear guidance for engineering decentralized, scalable IIoT defenses.