A Feature Selection Method for Intrusion Detection Based on Parallel Sparrow Search Algorithm
Hongwei Chen, Xin Ma, Song Huang · 2021
Aiming at the problems that traditional feature selection methods are difficult to optimize and have high computational complexity, this paper proposes a feature selection algorithm: Spark-based Improved Sparrow Search Algorithm (SPISSA). SPISSA is used to search feature subsets on intrusion detection data sets to obtain better classification accuracy. The standard sparrow search algorithm has the shortcomings of random initialization, fast convergence in the early iteration, and easy movement to the origin. In the SPISSA proposed in this paper, reverse learning and Levy flight random step are adopted, which improves the diversity and quality of sparrows in the initialization stage and enhances the global search ability in the later iteration stage. Finally, considering the high-dimensional and large-scale characteristics of network intrusion traffic, the algorithm is combined with Spark distributed computing framework. In spark framework, the population is calculated in parallel according to data partition. Experiments have proved that SPISSA can effectively find the optimal subset on the public data set. At the same time, the calculation time cost of the algorithm has been effectively reduced.