RT info:eu-repo/semantics/article T1 Comparison of Network Intrusion Detection Performance Using Feature Representation A1 Pérez, Daniel A1 Alonso Castro, Serafín A1 Morán Palao, Antonio A1 Prada Medrano, Miguel Ángel A1 Fuertes Martínez, Juan José A1 Domínguez, Manuel A2 Ingenieria de Sistemas y Automatica K1 Ingeniería de sistemas K1 Anomaly detection K1 Feature representation K1 Network intrusion detection AB Intrusion detection is essential for the security of the componentsof any network. For that reason, several strategies can be used inIntrusion Detection Systems (IDS) to identify the increasing attempts togain unauthorized access with malicious purposes including those baseon machine learning. Anomaly detection has been applied successfully tonumerous domains and might help to identify unknown attacks. However,there are existing issues such as high error rates or large dimensionalityof data that make its deployment di cult in real-life scenarios. Representationlearning allows to estimate new latent features of data in alow-dimensionality space. In this work, anomaly detection is performedusing a previous feature learning stage in order to compare these methodsfor the detection of intrusions in network tra c. For that purpose,four di erent anomaly detection algorithms are applied to recent networkdatasets using two di erent feature learning methods such as principalcomponent analysis and autoencoders. Several evaluation metrics suchas accuracy, F1 score or ROC curves are used for comparing their performance.The experimental results show an improvement for two of theanomaly detection methods using autoencoder and no signi cant variationsfor the linear feature transformation PB Springer SN 1865-0929 LK http://hdl.handle.net/10612/11293 UL http://hdl.handle.net/10612/11293 NO International Conference on Engineering Applications of Neural Networks, 2019 NO P. 463-475 DS BULERIA. Repositorio Institucional de la Universidad de León RD 29-mar-2024