<p>The Medical Internet of Things (IoT) has brought healthcare into the 21st century by connecting medical devices and allowing real-time patient monitoring. The tech has gotten a lot better, but it has also left its share of cybersecurity vulnerabilities that allow sophisticated and evolving cyber threats to compromise critical patient data. To address these challenges, this study proposes a Data-Driven Cognitive Cyber-Physical System (C-CPS) for efficient cyber threat mitigation in MIoT networks. The C‑CPS combines a hybrid approach of Convolutional Neural Networks (CNN) for feature extraction and Long Short-Term Memory (LSTM) networks for temporal anomaly detection, achieving high accuracy. We propose the use of the Recursive Feature Elimination (RFE) technique for feature selection to improve the efficiency of the system, by reducing the computational overhead by selecting the most important features. We optimize the system performance with the Bat Search Optimization Algorithm (BSOA) to fine-tune the hyperparameters of the CNN LSTM model. The proposed system achieved a detection accuracy of 98.2%, AUC-ROC score of 0.982, and outperformed the baseline models. Additionally, RFE reduced the input feature set by 40% without loss of accuracy, thus increasing model efficiency. Finally, the C CPS framework, based on BSOA, CNN LSTM, and RFE, can solve cyber threat mitigation in the MIoT environment robustly and efficiently.</p>