Adaptive Uncertainty-Aware CNN-BiLSTM and Neutrosophic Petri Net Framework for Real-Time IDS/IPS
Journal of Al-Farabi for Engineering Sciences · 2025
DOI: 10.59746/4cdvge88 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Al-Farabi for Engineering Sciences, 2025. Verfasst von Hind Ali Abdul Hassan.
Abstract (Originalsprache)
A crucial function of intrusion detection and prevention systems (IDS/IPSs) is to safeguard corporate, cloud, and cyber-physical networks against ever-changing, intricate threats. Class imbalance, high false-positive rates, poorly calibrated confidence estimates, limited interpretability, and strict prevention mechanisms that force uncertain traffic into binary allow-or-block decisions are still issues with many current IDS/IPS solutions, even though deep learning models can successfully capture nonlinear and temporal traffic patterns. An attention mechanism, conformal prediction, an Adaptive Neutrosophic Petri Net, convolutional neural networks, bidirectional long short-term memory networks, and an adaptive uncertainty-aware intrusion detection system/intrusion prevention system are all part of the proposed architecture in this research. In order to detect ambiguous and out-of-distribution traffic, the CNN-BiLSTM-Attention component extracts bidirectional temporal correlations and local feature interactions, while conformal prediction creates calibrated prediction sets. The preventive layer is able to route traffic to allow, block, quarantine, or deep-inspection actions based on the resultant probabilities, conformal ambiguity, entropy, evidence of traffic-drift, and contextual risk, which are mapped into adaptive truth, indeterminacy, and falsity values. The framework was assessed utilizing random, time-based, leakage-free, and cross-dataset protocols on the CSE-CIC-IDS2018 and comparable NetFlow based datasets. In the random-split situation, the suggested model attained an accuracy of 99.84%, a macro-F1 of 99.31%, an MCC of 0.996, a false positive rate of 0.13% and a false negative rate of 0.21%. Under more strict leakage-free assessment, it achieved 99.06% accuracy and 98.14% macro-F1. Under cross-dataset testing, it achieved 96.72% accuracy and 94.06% macro-F1, which shows increased generalization under distribution shift. The full-framework improved the false-positive rate from 0.36% to 0.13% and the false-negative rate from 0.78% to 0.21% compared to the baseline CNN-BiLSTM-Attention. Conformal prediction lowered the estimated calibration error from 0.071 to 0.013, achieved 95.2% empirical coverage for a nominal coverage level of 95%, and enhanced the quarantine response for unknown or out-of-distribution traffic from 42.0% to 91.8%. The Adaptive Neutrosophic Petri Net further lowered the prohibited benign traffic to 1.7% and permitted the malicious traffic to 0.29% with an average decision delay of 2.03 ms per traffic window. Results show that the suggested framework increases detection performance, calibration, operational safety, uncertainty management, and decision-level explainability, thereby constituting a potential architecture for trustworthy real-time intrusion prevention.
Zitieren
APA
Hassan, H. A. A. (2025). Adaptive Uncertainty-Aware CNN-BiLSTM and Neutrosophic Petri Net Framework for Real-Time IDS/IPS. Journal of Al-Farabi for Engineering Sciences. https://doi.org/10.59746/4cdvge88
Vancouver
Hassan HAA. Adaptive Uncertainty-Aware CNN-BiLSTM and Neutrosophic Petri Net Framework for Real-Time IDS/IPS. Journal of Al-Farabi for Engineering Sciences. 2025. doi:10.59746/4cdvge88
BibTeX
@article{Hassan2025,
author = {Hassan, Hind Ali Abdul},
title = {Adaptive Uncertainty-Aware CNN-BiLSTM and Neutrosophic Petri Net Framework for Real-Time IDS/IPS},
journal = {Journal of Al-Farabi for Engineering Sciences},
year = {2025},
doi = {10.59746/4cdvge88}
}
EUVIMED