Autoencoder-based network anomaly detection

Abstract

Anomaly detection is critical given the raft of cyber attacks in the wireless communications these days. It is thus a challenging task to determine network anomaly more accurately. In this paper, we propose an Autoencoder-based network anomaly detection method. Autoencoder is able to capture the non-linear correlations between features so as to increase the detection accuracy. We also apply the Convolutional Autoencoder (CAE) here to perform the dimensionality reduction. As the Convolutional Autoencoder has a smaller number of parameters, it requires less training time compared to the conventional Autoencoder. By evaluating on NSL-KDD dataset, CAE-based network anomaly detection method outperforms other detection methods.