OPTIMIZED FEATURE SELECTION OF CYBERATTACK DETECTION IN HEALTHCARE IOMT DEVICES USING THE MULTI-PROTOCOL DATASET BASED ON ADVANCE MACHINE LEARNING METHOD
Kshitiz Agarwal, Sandhya Sharma
Department of Electronics and Communication Engineering,
Suresh Gyan Vihar University, Jaipur, India
Abstract. The growing integration of Internet of Medical Things (IoMT) devices in healthcare has brought important benefits for patient monitoring and medical data management, but it has also created new risks of cyberattacks. The current study applies statistical methods and feature selection techniques to the CICIoMT2024 multi-protocol dataset to understand attack behaviors and identify effective attributes for detection. The dataset contains 18 different types of attacks, grouped into classes such as distributed denial of service (DDoS), denial of service (DoS), reconnaissance, spoofing, and MQTT-based threats. Training data shows more than 1.6 million DDoS-UDP flows and more than 1.5 million DDoS-ICMP flows, while some categories like ping sweep and vulscan remain highly under-represented, reflecting a strong imbalance in attack distribution. Exploratory data analysis confirmed that 45 numeric features are present without missing values. Statistical tests such as Shapiro–Wilk and Kolmogorov–Smirnov showed that most variables do not follow a normal distribution, indicating non-linear behavior in network traffic. Kruskal–Wallis testing revealed significant differences in features such as Header_Length and Duration across classes. Correlation heatmaps highlighted strong dependencies between traffic indicators, underlining the importance of dimensionality reduction. Feature importance analysis from tree-based models identified IAT, Srate, Rate, fin_count, and Header_Length as the most discriminative attributes. These indicators capture variations in packet timing, session rate, and header properties that strongly separate attack categories from benign traffic. The findings highlight the effectiveness of combining statistical analysis with feature selection for enhancing IoMT security research. The outcome provides guidance for designing machine learning models that can handle imbalanced datasets while improving detection accuracy for healthcare device networks. Five different machine learning methods were selected for the present study for advance analysis which hwere following and have high level accuracy for this dataset.
VOLUME 10 ISSUE 02 2026: 1 – 12