Epistemic uncertain computing for intrusion detection with explainability & multi-criteria optimization using AA-NLS...

Area of Science:

• Computer Science
• Cybersecurity
• Artificial Intelligence

Background:

• Network intrusion detection systems (NIDS) face uncertainty due to incomplete data and unaddressed out-of-order packet arrivals, leading to false positives.
• Existing NIDS models often overlook the epistemic uncertainty introduced by out-of-order packet sequences, limiting detection accuracy.
• Accurate network security relies on robust intrusion detection that can effectively handle data uncertainties.

Purpose of the Study:

• To propose a novel framework for network intrusion detection that specifically analyzes and mitigates uncertainty from out-of-order packet arrivals.
• To enhance the accuracy and reliability of NIDS by incorporating advanced uncertainty quantification techniques.
• To develop a comprehensive system for detecting, explaining, and prioritizing network intrusions.

Main Methods:

• Implemented an Algebraic Average-based Neutrosophic Logic System (AA-NLS) for out-of-order packet uncertainty analysis.
• Utilized Weighted Distance Error Function-based K-Nearest Neighbour (WDEFKNN) for missing data imputation and Density-Based Beta Distribution Soergel Spatial Clustering of Applications with Noise (DBBDSSCAN) for pattern grouping.
• Employed a Generalized Riccati Uniform Scaling Orthogonal-based Gated Recurrent Unit (GRUSO-GRU) for intrusion detection and a Pareto Entropy-based SHapley Additive exPlanation (PESHAP) model for explainability.

Main Results:

• The proposed AA-NLS based framework achieved a high intrusion detection accuracy of 99.299% in a full multi-class setting.
• Demonstrated superior performance compared to prevailing Deep Learning (DL)-based NIDS approaches.
• Successfully analyzed epistemic uncertainty from out-of-order packets and temporal patterns using AA-NLS and Hidden Laplace Witten Bell Markov Model (HLWBMM).

Conclusions:

• The novel AA-NLS framework effectively addresses out-of-order packet uncertainty, significantly improving network intrusion detection accuracy.
• The integrated approach provides explainable AI for intrusion detection and prioritizes alerts using multi-criteria optimization.
• This research offers a robust and accurate solution for modern network security challenges.