الفهرس 
SummaryOnly 14 pages are availabe for public view
 |  | from | 175 |  |  |
| | | from | 175 | | |
Abstract
Over the past few years, intrusion protection systems (IPS) have drawn a mature research area in the field of computer networks. With the massive increase in system traffic, the problem of excessive and unrelated features has a significant impact on the rate of intrusion detection performance in the intrusion protection systems. These unimportant features affect the network classification in several aspects, namely a) slowing down the classification process in the protection systems, and b) preventing the classifier from making accurate decisions. To classify network traffic, whether harmful or normal, the use of machine learning algorithms has been applied in many previous research. Feature selection is a very important part of machine learning systems. Therefore, to obtain the accuracy and speed of performance for intrusion protection systems, we must reduce the dimensionality of the data used. An improved model for feature selection is proposed in this thesis to increase the accuracy of intrusion detection systems, hence improve the performance of an intrusion protection system. We evaluate the perfo rmance of the proposed model based on a comparison of several known algorithms as well as other algorithms that use deep learning with a multi-layered perception algorithm. The NSL-KDD and UNSW-NB15 datasets have been used for examining classification capability of the intrusion protection system. Comparative study showed that the detection accuracy by UNSW-NB15 dataset is better than NSL-KDD dataset. The proposed model outperformed the other learning approaches referred to the references in many significant performance indicators.