Research Article
Wireless Local Area Network Intrusion Detection System Using Deep Belief Networks
@INPROCEEDINGS{10.1007/978-3-030-80621-7_23, author={Temesgen Mihiretu Abebe and Menore Tekeba Mengistu}, title={Wireless Local Area Network Intrusion Detection System Using Deep Belief Networks}, proceedings={Advances of Science and Technology. 8th EAI International Conference, ICAST 2020, Bahir Dar, Ethiopia, October 2-4, 2020, Proceedings, Part I}, proceedings_a={ICAST}, year={2021}, month={7}, keywords={Wireless Intrusion Detection System (WIDS) AWID dataset Deep Belief Networks (DBN) Feature selection}, doi={10.1007/978-3-030-80621-7_23} }- Temesgen Mihiretu Abebe
Menore Tekeba Mengistu
Year: 2021
Wireless Local Area Network Intrusion Detection System Using Deep Belief Networks
ICAST
Springer
DOI: 10.1007/978-3-030-80621-7_23
Abstract
In computer security Intrusion Detection System (IDS) is a mechanism of detecting an intruder in the system and notifying malicious activities for system administrators. The IDS researches on wireless Local Area Network (LAN) started recently. Until now there are some researches like publishing Aegean Wi-Fi Intrusion Dataset (AWID) dataset publically for the research community and evaluating the dataset using different machine learning algorithms. In this paper, we propose Deep Belief Network (DBN) to evaluate AWID dataset for intrusion detection analysis. Since AWID dataset contains different data types which are numeric, string, and hexadecimals; before training the model and evaluation of its performance the dataset is preprocessed and finally 102 attributes are used for system training. Also, two-stage feature selection is implemented to reduce the training cost and improve system performance. The first stage is removing duplicated attributes which reduced the dataset size to 68 attributes. The second stage is done by applying Weka implemented Information Gain Ratio (IGR). Using three thresholds three datasets are prepared with 41 attributes, 34 attributes, and 25 attributes. The system was able to achieve 98.55% accuracy with 102 attributes and it was able to improve this result to 98.97% with selected 34 attributes set evaluation.
