A Novel Method to Enhance the Kernel Matrix of Support Vector Machine Classifier Using Genetic Algorithm for Intrusion Detection System

Support vector machine (SVM) is a notable classification algorithm in the application of intrusion detection system (IDS). It achieves high accuracy in categorizing network traffic as normal or attack data. However, it has a drawback of high misclassification rates in noisy environments. To address this, wrapper-based feature selection methods are employed to improve the classifier's performance. While these methods enhance the detection ratio by selecting the most informative features, they often suffer from local optima problems. In this paper, a novel method is proposed to enhance the performance of the SVM classifier by overcoming this limitation. The approach leverages a genetic algorithm (GA), matrix based vector representation (MBVR) and a random function to optimize the support vectors in the kernel matrix of a trained SVM classifier, instead of selecting informative features for training. The proposed method is evaluated using standard and simulated datasets. Results demonstrate that the enhanced kernel matrix improves the detection ratio by approximately 6% (CICIDS(L): 80%, ADFA-LD(L): 90%, 5G-NIDD: 99%) in large, comple compared to traditional GA-based feature selection methods of simple SVM, Logistic regression, ANN, naive bayes and decision tree classifiers. Additional analysis of random value ranges, computational, and communication complexity underscores the efficiency of the proposed SVM-based IDS.