Deep Learning-based Feature Selection for Intrusion Detection System in Transport Layer

Numerous machine learning algorithms applied on Intrusion Detection System (IDS) to detect enormous attacks. However, it is difficult for machine to learn attack properties globally since there are huge and complex input features. Feature selection can overcome this problem by selecting the most important features only to reduce the dimensionality of input features. We leverage Artificial Neural Network (ANN) for the feature selection. In addition, in order to be suitable for resource-constrained devices, we can divide the IDS into smaller parts based on TCP/IP layer since different layer has specific attack types. We show the IDS for transport layer only as a prove of concept. We apply Stacked Auto Encoder (SAE) which belongs to deep learning algorithm as a classifier for KDD99 Dataset. Our experiment shows that the reduced input features are sufficient for classification task. 한국정보보호학회 하계학술대회 논문집 Vol. 26, No. 1