Towards Building an Improved IoMT Attack Classification Model-A Methodological Framework and Analysis of the Experimental Dataset
Keywords:
Feature Selection, Model Performance, Internet of Things, Smart HealthcareAbstract
Internet of Medical Things (IoMT) is currently revolutionaising the healthcare industry as it can help in the advancement of peoples’ health condition as well as productivity of health professionals in service delivery. Despite the positive developments in the use of technology in health sector, people with malicious intent have been launching attacks on heterogeneous medical devices and networks. Machine learning (ML)-based models have been proven to have the ability to intelligently analyze huge volume of IoMT traffic and then classify attacks. To effectively detect zero-day attacks in IoMT with reduce rate of false positives using of ML-based methods, there is a need to address different issues such as high class imbalance, feature selection, and multi-class classification in experimental datasets. This study proposes a multi-stage methodological framework that can be used to adequately classify multi-class cyber-attacks using a recent CICIoMT2024 security dataset. The framework details the various stages that the proposed ML-based attack classification models will contain by taking some of the issues identified in the exploratory data analysis (EDA) layer into considerations. Lastly, Aquila Optimizer is suggested to be used for the selection of most promising attributes while Tree and Non-Tree based ensemble learning algorithms are proposed for the attack classification task. It is concluded that this methodological framework can be used for building improved multi-class attack classification models in IoMT scenario.
