Feature Selection for Robust Backscatter DDoS Detection

This paper analyzes the effect of using different feature selection algorithms for robust backscatter DDoS detection. To achieve this, we analyzed four different training sets with four different feature sets. We employed two well-known feature selection algorithms, namely Chi-Square and Symmetrical Uncertainty, together with the Decision Tree classifier. All the datasets employed are publicly available and provided by CAIDA. Our experimental results show that it is possible to develop a robust detection system that can generalize well to the changing backscatter DDoS behaviours over time using a small number of selected features.