EDGE-ENABLED MACHINE LEARNING FRAMEWORK FOR REALTIME ANOMALY DETECTION IN IOT NETWORK

Abstract

With the rapid expansion of IoT networks, global internet-connected devices are projected to exceed 29 billion by 2030, creating unprecedented volumes of real-time traffic. Nearly 70% of IoT devices are vulnerable to at least one security threat, and over 60% of reported anomalies go undetected due to inadequate early-warning mechanisms. The economic impact of network anomalies is significant, with businesses losing an estimated $120 billion annually due to undetected cyber threats, system downtime, and performance degradation. Traditional manual anomaly detection methods, such as signature-based identification, threshold monitoring, and log inspection, are increasingly ineffective in dynamic IoT environments. These techniques are time-intensive, prone to human error, and incapable of detecting zero-day anomalies or adapting to evolving traffic behaviors. To overcome these limitations, the proposed method presents a machine learning-driven anomaly detection framework tailored for IoT edge devices. The system begins with an end-to-end preprocessing pipeline that includes structured data exploration, class balance visualization, and feature standardization to optimize learning efficiency. Logistic Regression and an AdaBoost-enhanced Decision Tree Classifier are employed to classify four critical network anomaly types: Frequency Drift, Capacity Breach, Dual Signal Interference, and Request Overload. A performance evaluation module computes key metrics such as accuracy, precision, recall, and F1-score, and uses confusion matrices for interpretability. The architecture also supports real-time predictions on new data inputs, making the system practical for deployment in IoT-enabled infrastructures. By combining model reusability with scalable preprocessing and automated classification, this method enhances anomaly detection reliability and responsiveness in modern edge environments