标题：CYBER SECURITY BASED MACHINE LEARNING ALGORITHMS APPLIED TO INDUSTRY 4.0 APPLICATION CASE: DEVELOPMENT OF NETWORK INTRUSION DETECTION SYSTEM USING HYBRID METHOD
期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2020
卷号：98
期号：12
页码：2078-2091
出版社：Journal of Theoretical and Applied
摘要：The implementation of industry 4.0 is a complex process that involves several steps and management of a project involving all the company's resources: production, supply chain, engineering, maintenance, human resources, information systems and many others. faced of the risks weighing on companies, and even more on their sensitive data, the need to create a cybersecurity strategy for the Industry 4.0 is more essential than ever. It has been apparent for several years that industrial systems are vulnerable to computer attacks. This can be explained because they were not designed with security constraints, particularly because of their physical isolation from Internet. However, they now face a variety of attackers with different objectives and abilities. In this paper, we present our strategy of cybersecurity, based on machine learning algorithms, applied in the context of industry 4.0. For this purpose, Unified Threat Management based on machine learning algorithms (ML-UTM) was used. to touch all layers of pyramid of Computer Integrated Manufacturing we propose to put an ML-UTM between layer 4 and 3, and an Industrial Unified Threat Management based on machine learning algorithms (ML-IUTM) between layer 3 and 2, another ML-IUTM between layer 2 and 1. Then we will cite the works based on the use of filtering device between layers 1 and 0. This paper describes a machine learning approach to build an efficient and accurate network intrusion detection system, which is one of the features of UTM, using a hybrid method. Thus, we have combined the different machine learning algorithms, namely Support vector machine (SVM), One rule (OneR), K-nearest neighbor (K-NN) and Random forest (RF) with Particle Swarm Optimization (PSO) method using a real data set (Gas pipeline), and according to the results of our analysis, we have selected the best optimized classifier. The experimental results have demonstrated the reliability and efficiency of the proposed approach. The PSO method can provide various advantages to K-NN and RF classifiers such as higher accuracy, lower MSE and faster time to build model. After analyzing and comparing all these results it was found that the NIDS based RF optimized by PSO give the best performances, with accuracy of 99.30%, F-measure of 99,30% and MSE that has been reduced to 0,0034.
