基于萤火虫算法的特征选择方案

In intrusion detection, processing high-dimensional data is often an important step. High-dimensional data brings higher computing cost and more difficult to process. Therefore, it is essential to set up dimensionality reduction scheme. Aiming at the problems of easy convergence and high computational cost of the traditional firefly algorithm, this paper proposes a feature selection scheme based on the firefly algorithm by improving the firefly algorithm. The specific improvement methods are as follows: for the initial population, define a vector composed of 0-1 for a single firefly individual, and set a one-to-one correspondence for discrete features; For the step size, the dynamic update step size mechanism is introduced, and the initial step size is set to a higher value, which can better global search in the early stage of the algorithm iteration. After a certain number of iterations, the step size gradually decreases, and then quickly converges to the optimal solution; For the brightness function and selection mode, the distance between classes is introduced as the brightness function to redefine the selection mode. The improved firefly algorithm selects the features of KDDCUP99 data set and obtains 10 optimal features, which are \{3,4,5,6,8,10,13,23,24,37\}. After training the 10 features, the accuracy is 99.48\%.