ANOMALY- BASED INTRUSION DETECTION USING GENERATIVE ADVERSERIAL NETWORKS

Abstract

Intrusion detection system (IDS) has become vital role in the field of IT Security due to cyber security safety in all human and machine pass through day to day activities. Intrusion detection methods based on the signature-based techniques have been used widely with limitation of identify new emerging threats. However, the progress of technology and the shortcomings of the intrusion detection system are influenced to upgrade IDS based on signature. Anomaly-based IDS are to establish a normal behavior profile and then define abnormal behaviors by their degree of abnormality from the normal profile. One of the techniques is used algorithms that support Deep Learning. Generative Adversarial Networks (GANs) have been widely studied and applied in anomaly detection within 6 years from first introduced in 2014 due to their advanced advantage in generating and learning higher-dimensional data which is had high number of features such as images, sounds and text. On this paper we had use current existing GAN and WGAN one of GAN variants for anomaly intrusion detection using NSL KDD dataset. On the training phase we have used pre-processed data fed to algorithms to train with default parameters that the classification model is build. On the validation phase we have considered of loss and accuracy of each batch of data training through with optimal parameters that gather from grid search over cross validation. Finally, the selected trained model is used to predict the test dataset. The evaluation result showed that the accuracy in classifying normal and attack. The results had shown on WGAN with accuracy of 89% prediction with default parameter and high prediction that performing with accuracy of 95.7% with optimized parameter.