The Internet of Medical Things (IoMT) connects medical devices like heart monitors and insulin pumps to the internet, enhancing healthcare but exposing it to cyber attacks such as Distributed Denial-of-Service (DDoS), Spoofing, and Reconnaissance. This paper proposes a novel ensemble machine learning based cyber attack detection framework to secure IoMT networks from several attacks including rare ones such as spoofing and reconnaissance. Our detection system as proposed in the paper first employ a Convolutional Neural Network (CNN) with Local Interpretable Model-Agnostic Explanations (LIME) to filter irrelevant features from the well-known CICIoMT2024 dataset. Then, a novel ensemble learning based methodology for efficient detection of rare attacks is proposed which combines XGBoost, Random Forest, and LightGBM together and uses Optuna for hyperparameter optimization. Three derived features are also incorporated by our model to enhance system’s adaptability for covering rare attacks in IoMT networks.