Network intrusion detection using stacked denoising autoencoder

The packets used in network intrusion detection contain noises and outliers. So, when the attacks are detected, it causes performance degradation. Therefore, to improve the performance of the intrusion detection system, it is necessary to remove the noise and outliers in the network packet. The autoencoder is an unsupervised learning model that reconstructs the input data at the output layer. In the process of reconstruction, the autoencoder removes the noise or outliers in the input data by repeating the encoding and decoding and reduces the dimensions for the input data by using latent variable in the hidden layer. Therefore, data reconstruction by the autoencoder allows it to obtain the data from which noise and outliers are removed, which in turn eliminates the negative effects on training. In this paper, we make the Stacked Denoising Autoencoder (SdA) learn the KDD Cup 1999 datasets with added noise. And then we remove the noise and outliers contained in the input data by using the learned SdA and input the reconstructed data into the intrusion detection system. As a result, it was found that when there are noise and outliers in the input data, it is possible to prevent the degradation of network intrusion detection model performance by reconstructing the input data through learned SdA to remove the noise and outliers.