This paper introduces SmartEdge, an advanced framework designed to enhance the processing and analysis of Internet of Medical Things (IoMT) data. SmartEdge combines systematic preprocessing, comprehensive feature selection, and an innovative model aggregation approach to minimize computational overhead, making it highly suitable for resource-constrained edge devices. Its adaptive learning capabilities enable ongoing optimization in response to new IoMT data and dynamic operational conditions. The study demonstrates the effectiveness of feature engineering on the CICIoMT2024 dataset, achieving a substantial reduction in data dimensionality through methods such as Principal Component Analysis (PCA) and embedding techniques. These strategies lower computational loads by up to 95%, facilitating real-time data processing and enabling the deployment of machine learning models directly on edge devices. SmartEdge achieves rapid training of 10 edge models in just 55 s, maintaining an accuracy rate above 0.99 for cyber attack detection. These results highlight the framework’s potential to significantly enhance IoMT security, resilience, and operational efficiency, making it a vital contribution to the development of secure and efficient IoMT ecosystems.