| Abstract |
The Internet of Things (IoT), which has had a revolutionary influence on human existence, has become a topic of significant attention among the scientific and industrial communities. Smart healthcare, smart cities, smart devices, smart industry, smart grid, and smart cities are just a handful of the many IoT ideas that have altered human life due to the rapid progress of this IoT technology. Security issues involving IoT devices have come up as a significant issue in recent years with special emphasis on the healthcare sector. This increased emphasis is mostly due to the exposure of serious vulnerabilities in IoT security with recent hacking activities. There is significant proof that conventional methods of protecting networks are effective. Still, the use of conventional security protocols for protection of IoT gadgets and networks from hacking is not feasible due to the constrained resources associated with IoT devices and the distinct characteristics observed in IoT protocols. To improve the privacy of the IoT, researchers will need a unique collection of resources, techniques, and datasets in IoT field. To address the earlier described issues, CatBoost is an innovative ensemble approach that combines many tree techniques and optimizes for performance. This model aims to accurately and automatically detect instances of assaults and anomalies in IoT sensors within the healthcare domain. For the successful creation of a security-based model, the hyperparameters are tuned with self-adaptive memetic firefly algorithm (SAMFA) optimization. The primary advantages of this study include (i) The development of an improved ensemble learning CatBoost model-based security system for IoT healthcare network intrusion detection, (ii) the SAMFA optimization method has been implemented for determining the ideal set of hyperparameters for the CatBoost algorithm, and (iii) Assessing the model's performance with a novel dataset of real-life observations (IoT Healthcare Security Dataset). The suggested model outperforms several previous state-of-the-art techniques, with experimental findings indicating outstanding anomaly identification accuracy of 99.99%. © 2023, The Author(s), under exclusive licence to Springer Nature B.V. |
