An effective feature engineering for DNN using hybrid PCA-GWO for intrusion detection in IoMT architecture

Abstract

The entire computing paradigm is changed due to the technological advancements in Information and Communication Technology (ICT). Due to these advancements, various new communication channels are being introduced, out of which the Internet of Things (IoT) plays a significant role. The Internet of Medical Things (IoMT) is a special category of IoT in which the medical devices communicate with each other for sharing sensitive data. These advancements help the healthcare industry to have better contact and care towards their patients. But they too have certain drawbacks since there are so many security and privacy issues like replay, man-in-the-middle, impersonation, privileged-insider, remote hijacking, password guessing, denial of service (DoS) attacks and malware attacks. When the sensitive data is being attacked by any of these attacks, there is a chance of losing the authorized data to the attacker or getting altered due to which the data is not available for the authorized users and customers. Machine learning algorithms are widely used in the Intrusion Detection System (IDS) for detecting and classifying the attacks at the network and host level in a dynamic manner. Many supervised and unsupervised algorithms have been designed by researchers from the area of machine learning and data mining to identify the reliable detection of an anomaly. However, the main challenge in the IDS models are changed in dynamic and random behavior of malicious attacks and designing a scalable solution that can handle this behavior. The rapid change in network behavior and the fast evolution of various attacks paved the way for evaluating various datasets that are generated over the years and to design different dynamic approaches. In this paper, a deep neural network (DNN) is used to develop effective and efficient IDS in the IoMT environment to classify and predict unforeseen cyberattacks. The network parameter are preprocessed, optimized and tuned by hyperparameter selection methods. A comprehensive analysis of experiments in DNN with other machine learning algorithms are compared on the benchmark intrusion detection dataset. Through rigorous testing, it has proved that the proposed DNN model performs better than the existing machine learning approaches with an increase in accuracy by 15% and decreases in time complexity by 32%, which helps in faster alerts to avoid post effects of intrusion in sensitive cloud data storage.