Boundary-Seeking GAN Approach to Improve Classification of Intrusion Detection Systems Based on Machine Learning Model

Abstract

An intrusion Detection System or IDS is a device of the computer used to observe network traffic and to analyze attacks coming from illegal users. IDS data is also processed using Machine Learning (ML) to see patterns of attacks. However, some results of IDS data processing using traditional ML techniques such as Random Forest, Artificial Neural Network, and Decision Tree, abbreviated RF, DT, and ANN are still not optimal. Therefore, we recommend the RF, DT, and ANN approaches in combination with the Boundary-seeking Generative Adversarial Network or BGAN to enhance the ability of ML techniques. BGAN is used to generate more variety of data to be used for further analysis. It uses the discriminator's estimated difference measure to decide how significant each sample should be. With BGAN, the decision boundary of the discriminator is closely related to the considerable weights. The results demonstrate that our suggested BGAN data model can considerably enhance the performance of RF, DT, and ANN. In the future, BGAN for ML could be improved for real-time deployment and analysis of intrusion detection.