Improved hybrid teaching learning based optimization-jaya and support vector machine for intrusion detection systems

• Main Author: Mohammad Khamees Khaleel, Alsajri
• Format: Thesis
• Language: English
• Published: 2022
• Subjects: Q Science (General); QA75 Electronic computers. Computer science
• Online Access: http://umpir.ump.edu.my/id/eprint/38456/1/Improved%20hybrid%20teaching%20learning%20based%20optimization-jaya%20and%20support%20vector%20machine%20for%20intrusion%20detection%20systems.ir.pdf; http://umpir.ump.edu.my/id/eprint/38456/

Most of the currently existing intrusion detection systems (IDS) use machine learning algorithms to detect network intrusion. Machine learning algorithms have widely been adopted recently to enhance the performance of IDSs. While the effectiveness of some machine learning algorithms in detecting certain types of network intrusion has been ascertained, the situation remains that no single method currently exists that can achieve consistent results when employed for the detection of multiple attack types. Hence, the detection of network attacks on computer systems has remain a relevant field of research for some time. The support vector machine (SVM) is one of the most powerful machine learning algorithms with excellent learning performance characteristics. However, SVM suffers from many problems, such as high rates of false positive alerts, as well as low detection rates of rare but dangerous attacks that affects its performance; feature selection and parameters optimization are important operations needed to increase the performance of SVM. The aim of this work is to develop an improved optimization method for IDS that can be efficient and effective in subset feature selection and parameters optimization. To achieve this goal, an improved Teaching Learning-Based Optimization (ITLBO) algorithm was proposed in dealing with subset feature selection. Meanwhile, an improved parallel Jaya (IPJAYA) algorithm was proposed for searching the best parameters (C, Gama) values of SVM. Hence, a hybrid classifier called ITLBO-IPJAYA-SVM was developed in this work for the improvement of the efficiency of network intrusion on data sets that contain multiple types of attacks. The performance of the proposed approach was evaluated on NSL-KDD and CICIDS intrusion detection datasets and from the results, the proposed approaches exhibited excellent performance in the processing of large datasets. The results also showed that SVM optimization algorithm achieved accuracy values of 0.9823 for NSL-KDD dataset and 0.9817 for CICIDS dataset, which were higher than the accuracy of most of the existing paradigms for classifying network intrusion detection datasets. In conclusion, this work has presented an improved optimization algorithm that can improve the accuracy of IDSs in the detection of various types of network attack.