Machine Learning-Driven Intrusion Detection for Next-Generation Information Security Systems
International Research Journal of Multidisciplinary Technovation · 19.05.2026 · 273-288
DOI: 10.54392/irjmt26317 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Research Journal of Multidisciplinary Technovation, online erschienen am 19.05.2026. Verfasst von Rabins Porwal und 6 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The proliferation of cloud, IoT, edge, and 5G infrastructures has dramatically expanded the attack surface of modern networks, while many existing intrusion detection systems (IDS) remain centralized, poorly interpretable, and brittle to concept drift and adversarial manipulation. Traditional machine learning-based IDS architectures demand centralization of raw data, offer limited decision transparency, degrade as traffic distributions evolve, and scale poorly to privacy-sensitive and resource-constrained deployments. In this paper, the Adaptive Explainable Federated Intrusion Detection System (AEF-IDS) is proposed, incorporating privacy-preserving federated learning, Kolmogorov-Smirnov (KS) test-based drift detection, differential privacy, adversarial robustness training, and multi-level explainability within a unified edge-oriented framework. Evaluated on three widely adopted benchmarks, namely NSL-KDD, UNSW-NB15, and CIC-IDS2018, AEF-IDS achieves detection accuracies of 96.74%, 93.92%, and 95.87%, false positive rates of 1.68%, 2.61%, and 2.19%, and AUC-ROC scores of 0.9781, 0.9573, and 0.9683, respectively. The system satisfies strict real-time performance requirements, achieving per-sample inference latencies of 47.3, 44.8, and 46.1 ms across the three benchmarks, all within the 50 ms operational threshold. AEF-IDS further demonstrates high resilience against white-box adversarial attacks, including FGSM, PGD, C&W, and DeepFool, maintaining a mean under-attack detection accuracy exceeding 88% across all evaluated datasets. Through federated optimization and KS-triggered adaptive retraining, the system effectively mitigates distributional shift while preserving local data sovereignty, and SHAP/LIME-based explanations provide both global and local attribution transparency for security analysts. These results collectively demonstrate that AEF-IDS constitutes a robust, privacy-preserving, and interpretable solution for next-generation IDS deployment at the network edge. Future work will address cross-domain generalization, online hyperparameter adaptation, and large-scale real-world field validation.
Zitieren
APA
Porwal, R., Adhikari, M. S., S, K., Yadav, A. K., Ketha, M. B., Verma, P., & Kunal (2026). Machine Learning-Driven Intrusion Detection for Next-Generation Information Security Systems. International Research Journal of Multidisciplinary Technovation, 273-288. https://doi.org/10.54392/irjmt26317
Vancouver
Porwal R, Adhikari MS, S K, Yadav AK, Ketha MB, Verma P, et al. Machine Learning-Driven Intrusion Detection for Next-Generation Information Security Systems. International Research Journal of Multidisciplinary Technovation. 2026:273-288. doi:10.54392/irjmt26317
BibTeX
@article{Porwal2026,
author = {Porwal, Rabins and Adhikari, Manoj Singh and S, Keerthi and Yadav, Anil Kumar and Ketha, Mahesh Babu and Verma, Piyush and Kunal},
title = {Machine Learning-Driven Intrusion Detection for Next-Generation Information Security Systems},
journal = {International Research Journal of Multidisciplinary Technovation},
year = {2026},
pages = {273--288},
doi = {10.54392/irjmt26317}
}
EUVIMED