Model selection for anomaly detection in wireless ad hoc networks

Anomaly detection has been actively investigated to enhance the security of wireless ad hoe networks. However, it also presents a difficulty on model determination, such as feature selection and algorithm parameter optimization. In this paper, we address the issue of parameter selection for one-class Support Vector Machine (1-SVM) based anomaly detection. We have investigated the performance of existing approaches, and also proposed a skewness-based outlier generation approach for parameter selection in the 1-SVM based anomaly detection model.