A feature selection approach to find optimal feature subsets for the network intrusion detection system

The performance of network intrusion detection systems based on machine learning techniques in terms of accuracy and efficiency largely depends on the selected features. However, choosing the optimal subset of features from a number of commonly used features to detect network intrusion requires extensive computing resources. The number of possible feature subsets from given n features is 2$$^{n}-1$$n-1. In this paper, to tackle this problem we propose an optimal feature selection algorithm. Proposed algorithm is based on a local search algorithm, one of the representative meta-heuristic algorithms for solving computationally hard optimization problems. Particularly, the accuracy of clustering obtained by applying k-means clustering algorithm to the training data set is exploited to measure the goodness of a feature subset as a cost function. In order to evaluate the performance of our proposed algorithm, comparisons with a feature set composed of all 41 features are carried out over the NSL-KDD data set using a multi-layer perceptron.