BRL-ETDM: Bayesian reinforcement learning-based explainable threat detection model for industry 5.0 network

To enhance the universal adaptability of the Real-Time deployment of Industry 5.0, various machine learning-based cyber threat detection models are given in the literature. Most of the existing threat detection models may not be able to detect zero-day cyber threats and are prone to producing a high...

Full description

Saved in:
Bibliographic Details
Published inCluster computing Vol. 27; no. 6; pp. 8243 - 8268
Main Authors Dey, Arun Kumar, Gupta, Govind P., Sahu, Satya Prakash
Format Journal Article
LanguageEnglish
Published New York Springer US 01.09.2024
Springer Nature B.V
Subjects
Access control
Accuracy
Algorithms
Artificial intelligence
Bayesian analysis
Computer Communication Networks
Computer Science
Cybersecurity
Datasets
Explainable artificial intelligence
Exploitation
Feature selection
Industry 4.0
Internet of Things
Machine learning
Methods
Operating Systems
Processor Architectures
Threats
Feature selection
Explainable AI
Industry 5.0
Cyber security
Reinforcement learning
Majority voting
Online AccessGet full text

Cover

More Information
Summary:To enhance the universal adaptability of the Real-Time deployment of Industry 5.0, various machine learning-based cyber threat detection models are given in the literature. Most of the existing threat detection models may not be able to detect zero-day cyber threats and are prone to producing a high False Positive Rate (F PR ) due to irrelevant features and imbalanced class samples. Furthermore, its predictive decisions are also difficult to comprehend even by security experts. Consequently, an intelligent and more robust model is needed to mitigate zero-day cyber threats. This study proposes an explainable model named BRL-ETDM for detecting cyber threats in Industry 5.0. In this model, features are optimized by Bayesian Reinforcement Learning ( BRL )-based Bee Swarm Optimization ( BSO ) technique in which the exploitation phase of BSO is improved by the BRL technique. Then, an improved weighted majority voting-based ensemble technique is designed to enhance threat detection performance. Additionally, an explainable AI technique is employed to explain the threat predictions. This model is tested and validated using two realistic datasets named Edge-IIoTset and ToN-IoT. Experimental results show that the proposed model achieved a maximum accuracy of 96.15% with a minimum number of features and F PR of 0.27% as compared to existing techniques.
ISSN:1386-7857
1573-7543
DOI:10.1007/s10586-024-04422-6