Smart intrusion detection system with balanced data in IoMT infra

Article type: Research Article

Authors: Umamaheswaran, S.^{a; *} | Mannar Mannan, J.^b | Karthick Raghunath, K.M.^c | Dharmarajlu, Santhi Muttipoll ^d | Anuratha, M.D.^e

Affiliations: [a] Department of Artificial Intelligence and Machine Learning, New Horizon College of Engineering, Bangalore, Karnataka, India | [b] Department of Computer Science and Engineering, EC Campus, PES University, Bangalore, Karnataka, India | [c] Department of Computer Science and Engineering, Jain University, Bangalore, Karnataka, India | [d] College of Nursing, Jazan University, Jazan, Kingdom of Saudi Arabia | [e] Professor cum Nursing superintendent, PSG Hospitals, Coimbatore, Affiliated with The Tamilnadu Dr MGR Medical University, Chennai, Tamilnadu, India

Correspondence: [*] Corresponding author. S. Umamaheswaran, Department of Artificial Intelligence and Machine Learning, New Horizon College of Engineering, Bangalore, Karnataka, India. E-mail: maheshumas@yahoo.com.

Abstract: The IoMT (Internet of Medical Things) has allowed for uninterrupted, critical patient observation, improved diagnosis precision, and efficient therapy. However, despite the usefulness of such medical things (devices), they also raise a lot of confidentiality and security issues since they provide potential entry points for hackers to exploit. Therefore, there is a pressing need for a technique for detecting network intrusions that combines precision, flexibility, and consistency. Addressing diverse information sources is challenging for finding negligible intrusions in sophisticated network systems, a core problem for current Intrusion Detection Systems (IDS). In this research, we propose a deep learning-based method for efficient network IDS in cases when data is unevenly distributed. Therefore, to address the poor identification rate of intrusions, we present a unique CGAN-CNN (Conditional Generative Adversarial Network-Convolutional Neural Network) IDS approach that oversamples from the unbalanced information based on the CGAN paradigm to overcome the functional deterioration induced by such unbalanced data, especially during intrusion detection. In addition, the sub-networks’ critic and generator each get additional constraints as part of the CGAN’s standard operating procedure, which helps to reduce the amount of leeway in the convergence process and speeds up the impact of convergence. To validate the effectiveness of the suggested model, we conducted an investigation using the most contemporary publicly available datasets, namely NIDS (Network Intrusion Detection System), and the CICDDoS2019 (Canadian Institute for Cybersecurity Distributed Denial of Service 2019) dataset from the Canadian Institution for Cybersecurity, and for healthcare-oriented image datasets Kaggle, respectively. The experimental findings validated the superiority of the CGAN-CNN approach described in this research. Notified as more trustworthy indications, F1-score and precision performed at 97.88%, and 97.15%, respectively.

Keywords: Deep learning, IDS, dataset, balancing, attacking, intrusions, generator, discriminator, CGAN, detection, evaluation

DOI: 10.3233/JIFS-233649

Journal: Journal of Intelligent & Fuzzy Systems, vol. 46, no. 2, pp. 3191-3207, 2024