AnoML-IoT: Adaptive Ensemble Learning Framework for Intelligent Anomaly Detection in IoT Sensor Networks

The rapid growth of Internet of Things (IoT) applications has increased serious security challenges, including system vulnerabilities, inefficient resource usage, and risks to real-time data integrity in heterogeneous sensor networks. Traditional anomaly detection techniques and single machine-learning models struggle with class imbalance, scalability, and low detection accuracy in distributed IoT environments. To overcome these limitations, this paper proposes AnoML-IoT, an adaptive multi-layer ensemble learning framework for intelligent anomaly detection. The framework combines Random Forest, Gradient Boosting, Decision Tree, and Isolation Forest models using an adaptive weighted voting mechanism supported by cross-validation. To handle class imbalance, the SMOTE-Tomek sampling method is employed, while feature engineering and edge-based preprocessing enhance data quality and processing efficiency. Experimental evaluation using real-world IoT sensor data demonstrates superior anomaly detection performance compared to conventional approaches. The system achieves 95.12% accuracy with a Support Vector Machine and 100% accuracy when deployed as an ensemble classifier. Feature importance analysis identifies temperature, time, humidity, and light as key indicators of anomalous behavior. The proposed framework maintains low false-positive rates, strong generalization capability, and effective scalability, enabling reliable real-time deployment. AnoML-IoT provides a robust and dependable anomaly detection solution for smart cities, industrial IoT, and smart healthcare systems. By strengthening the security, resilience, and reliability of IoT infrastructures, the proposed framework directly leads to the objectives of Industry, Innovation, and Infrastructure; Sustainable Cities and Communities; and Good Health and Well-Being as it allows to obtain secure, efficient, and trustworthy IoT-based services.