|
Deep learning and a cost-sensitive strategy for intrusion detection system
|
Maryam Pournagdi   |
|
|
|
Abstract: (910 Views) |
Intrusions and attacks are critical problems in network security and privacy. There are many studies on intrusion detection, most of which use traditional data mining algorithms to detect intrusion. Intrusion detection using deep learning is a new approach to cyber security. Unbalanced data is one of the major challenges in intrusion detection in which the number of samples of some classes (majority) is much higher than other classes (minority) which increases the rate of incorrect classifications for minority classes and the created model tends towards the majority class. Although some studies have tried to address this issue by using resampling techniques, they are not effective. This research uses deep learning that combines the phases of classification and automatic feature extraction. Unlike previous approaches, the proposed method addresses the class imbalance problem during model training. The developed mechanism uses a cost-sensitive learning strategy and determines a cost for each misclassification based on the class distribution. These costs are considered during training when there are incorrect classifications to fit the data. The efficiency of the designed approach is assessed using different criteria such as accuracy, recall, precision, and F1-Score by the same method. Experiments have shown that the proposed method can improve the classification performance by an average of 3%.
|
Article number: 6 |
| Keywords: Intrusion detection, Imbalanced data, Deep learning, Cost-sensitive learning |
|
|
Full-Text [PDF 1648 kb]
(326 Downloads)
|
Type of Study: Research |
Subject:
General
Received: 2023/01/1 | Accepted: 2023/01/1 | Published: 2023/01/1
|
|
|
|
|
|
|
| Add your comments about this article |
|
|
|
