ANOMALY DETECTION IN IOT NETWORKS USING AI MACHINE LEARNING AND STATISTICAL MODELS

Abstract

The explosive growth of the Internet of Things (IoT) has significantly expanded the attack surface across modern digital ecosystems, making efficient and accurate anomaly detection a critical requirement for ensuring system resilience. This research presents a hybrid anomaly detection framework that integrates supervised machine learning models, unsupervised deep-learning components, and statistical decision mechanisms to detect malicious behavior in IoT network traffic. Using the BoT-IoT dataset as an experimental benchmark, the system employs a multi-stage architecture involving data preprocessing, feature engineering, supervised classification with XGBoost, and anomaly scoring using an LSTM autoencoder. Comparative evaluation across multiple metrics including accuracy, precision, recall, F1-score, detection rate, false-positive rate, latency, and throughput demonstrates strong suitability for real-time IoT deployments.

Experimental results show that the proposed hybrid model significantly outperforms conventional baselines, with XGBoost achieving 99.4% accuracy and the LSTM autoencoder attaining a 97.6% detection rate alongside a low 2.1% false-positive rate. Statistical validation confirms the robustness and generalization capabilities of the architecture under varying traffic patterns, while latency and throughput evaluations indicate practical feasibility for deployment in resource-constrained environments. The results highlight the effectiveness of combining machine learning and statistical anomaly scoring to detect both known and unknown cyber threats in heterogeneous IoT networks, ultimately offering a reliable, scalable, and real-time security solution for modern IoT infrastructures.