An unsupervised anomaly detection engine with an efficient feature set for AODV

Abstract

There are some security issues in Mobile Ad hoc Networks (MANETs) due to mobility, dynamic topology changes, and lack of any infrastructure. In MANETs, it is of great importance to detect anomaly and malicious behavior. In order to detect malicious attacks via intrusion detection systems and analyze the data set, we need to select some features. Hence, feature selection plays critical role in detecting various attacks. In the literature, there are several proposals to select such features. Usually, Principal Component Analysis (PCA) analyzes the data set and the selected features. In this paper, we have collected a feature set from some state-of-the-art works in the literature. Actually, our simulation shows this feature set detect anomaly behavior more accurate. In addition, for the first time, we use robust PCA for analyzing the data set instead of PCA in MANET. By means of robust PCA, we have an unsupervised algorithm versus semi-supervised provided by PCA. In contrast to PCA, our results show robust PCA cannot be affected by outlier data within the network. In this paper, normal and attack states are simulated and the results are analyzed.