A Quantum Feature Selection Method for Network Intrusion Detection

Feature selection (FS) approaches rank features based on the score of the coefficients with labels. However, these selected features usually lead to a suboptimal solution to classification problems as they are selected independently. Moreover, with large dimensional feature sets, the computational complexity of FS algorithms can be prohibitively high. In this paper, we first formulate the feature selection problem as an integer programming model. Then we propose to utilize the strong computation ability of quantum annealers to solve the discrete integer programming problems, where the quantum annealer has the potential to be significantly faster than classical solvers to solve discrete optimization problems. We also design a wrapper algorithm to choose the optimal parameters of QUBO. Experiments show that our proposed strategy can select the representative features in the NSL-KDD dataset. Compared with HHO, WOA, PSO, and other algorithms, our strategy retains the least features to minimize the detection time, while the accuracy increase to 89.2%. Our algorithm also shows a good performance in computation time, detection rate and precision.