Authors:
Weijinxia
1
;
Longchun
2
;
1
;
Wanwei
2
;
1
;
Zhaojing
1
;
Duguanyao
2
;
1
and
Yangfan
1
Affiliations:
1
Department of Security, Computer Network Information Center, Chinese Academy of Sciences, Beijing, 100190, China
;
2
Chinese Academy of Sciences University, Beijing, 101408, China
Keyword(s):
Intrusion Detection, Class Imbalance, I-SMOTE, Feature Reduction, Random Forest.
Abstract:
With the wide applications of network in our daily lives, network security is becoming increasing prominent. Intrusion detection systems have been widely used to detect various types of malicious network which cannot be detected by a conventional firewall. Therefore, various machine-learning techniques have been proposed to improve the performance of intrusion detection system. However, the balance of different data classes is critical and will affect detection performance. In order to reduce the impact of class imbalance of the intrusion dataset, this paper proposes a scheme that applies the improved synthetic minority oversampling technique (I-SMOTE) to balance the dataset, employs correlation analysis and random forest to reduce features and uses the random forest algorithm to train the classifier for detection. The experimental results based on the NSL-KDD dataset show that it achieves a better and more robust performance in terms of accuracy, detection rate, false alarms and tra
ining speed.
(More)