| Abstrakt: |
Network intrusions have predominantly increased following the rapid expansion of networks in different areas such as social networking, e-learning, e-business, etc. With the rapid growth of network traffic, malicious users are inventing new methods of network intrusion. Therefore, intrusion detection systems (IDS) are developed to deal with such harmful activities. In fact, the huge amount of network data, which may contain irrelevant/redundant and noisy characteristics and features, has bombarded machine learning and data mining tools with unprecedented challenges. Hence, reducing the network data dimensionality can enhance the model interpretability, reduce resources requirements and increase generalizability. In this paper, an efficient ensemble feature selection for intrusion detection system (IDS-EFS) is proposed to select the best performing subset for attacks detection. The performances of IDS-EFS are compared with the well-known feature selection methods using the KDDCup-99 network dataset. The recorded results confirm the efficiency of our system in terms of accuracy, recall, precision, f1-measure, AUC score and running time. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink