Automatic Features Extraction Using Autoencoder in Intrusion Detection System

Abstract

Intrusion Detection System (IDS) can detect attacks by analysing the patterns of data traffic in the network. With a large amount of data that is processed in the IDS, then need to do a feature extraction to reduce the computational cost of processing raw data in IDS. Feature extraction will transform features to the lower dimension to accelerate the learning process and improve the accuracy. This research on automatic feature extraction using simple autoencoder and SVM to classify attacks on IDS. We use various functions activation and loss to see how far this feature extraction feature can improve accuracy. We use Datasets KDD Cup‘ 99 NSL-KDD and to evaluate the effectiveness of the mechanisms of detection after extraction features process. In the proposed model, the activation functions autoencoder hyperparameter ReLU activation and loss function cross-entropy gives best accuracy value than other functions.