Machine Learning-Driven QoS Optimization for IoT in OneM2M: A Novel Approach for Traffic and Resource Management

The rapid growth of IoT applications has highlighted the critical importance of efficient Quality of Service (QoS) management in platforms like OneM2M. This paper presents an innovative machine learning-based approach, applied for the first time to the OneM2M standard, to optimize QoS parameters under high data traffic scenarios. This study examines the application of sophisticated data balancing techniques alongside ensemble learning methods to improve the performance of classification models for traffic management in IoT environments. Our approach begins by addressing the inherent data imbalance in IoT-generated traffic, leveraging synthetic over-sampling techniques such as SMOTE. This ensures that classification models are not biased towards dominant traffic classes. Additionally, we assess the effectiveness of various machine learning models, such as Random Forest, Support Vector Machines, XGBoost, and LightGBM, using both imbalanced and balanced datasets. To achieve superior accuracy and robustness, we implement ensemble learning strategies, including Voting and Stacking Classifiers, combining the strengths of Random Forest and LightGBM. These ensemble methods demonstrate significant improvements in prediction accuracy, achieving a weighted F1-score of 98%. Our results reveal that machine learning can effectively classify and prioritize IoT traffic, thereby optimizing critical QoS metrics such as latency, success rate, CPU, and RAM utilization. By applying these models within OneM2M, we demonstrate their potential to intelligently manage traffic congestion, filter non-IoT traffic, and autonomously migrate excessive load to the cloud. This study sets a foundation for deploying ML-driven QoS management in real-world IoT systems and highlights its transformative potential for ensuring reliable, efficient, and scalable IoT services.