A stochastic data discrimination based autoencoder approach for network anomaly detection

Abstract

Machine learning based network anomaly detection methods, which are already effective defense mechanisms against known network intrusion attacks, have also proven themselves to be more successful on the detection of zero-day attacks compared to other types of detection methods. Therefore, research on network anomaly detection using deep learning is getting more attention constantly. In this study we created an anomaly detection model based on a deterministic autoencoder, which discriminates normal and abnormal data by using our proposed stochastic threshold determination approach. We tested our proposed anomaly detection model on the NSL-KDD's test dataset KDDTest+ and obtained an accuracy of 88.28%. The experimental results show that our proposed anomaly detection model can perform almost the same as the most successful and up-to-date machine learning based anomaly detection models in the literature.