A Novel Meta-Heuristic-Based Sequential Forward Feature Selection Approach for Anomaly Detection Systems
Yukang Liu, Zhen Xu, Jing Yang, Liming Wang, Chen Song, Kai Chen · 2016
Anomaly detection technique play an extraordinary role in the Intrusion Detection System (IDS) for its ability to detect novel attacks. To overcome the high-dimensionality problem the anomaly detection cursed of, we propose a novel Meta-Heuristic-based Sequential Forward Selection (MH_SFS) feature selection algorithm, which can be generally implemented in anomaly detection system. It is an improvement of the traditional Sequential Forward Selection (SFS) feature selection algorithm. After every iteration of SFS, we append a meta-heuristic search to tackle the "nesting effect" problem. And a filter-based function is added to counteract the computational cost from the additional metaheuristic search process. During the testing phase, based on KDD Cup 99 dataset, we compare our improved algorithm to the traditional SFS algorithm and the state-of-the-art IFFS algorithm over three anomaly detection models. The experimental results strongly show that our algorithm can yield higher detection performance and smaller number of features selected, while computations are in a permitted range.