基于元学习和二进制粒子群的网络故障特征选择算法

A Meta-learning and Binary Particle Swarm Optimization Based Network Fault Feature Selection Algorithm

为了降低Wrapper模式网络故障特征选择方法分类算法的计算量,文章提出了一种基于元学习和二进制粒子群(ML-BPSO)的特征选择方法;算法在封装的分类训练中采用元学习方法估算分类精度,并利用BPSO在特征空间中进行全局搜索选出最优特征集;在DARPA数据集上的实验可以看出本文方法选取结果与BPSO-SVM相当但是计算量大大降低;实验结果表明文章提出的方法能够显著的降低网络故障特征选择计算量,同时保证了较高的诊断精度和较好的降维效果。

The wrapper network fault feature selection algorithms get large calculation cost,a Meta-learning and binary particle swarm optimization （ML-BPSO） based feature selection algorithm was proposed to solve this problem in this paper.The Meta-learning method was introduced for estimating the classification accuracy wrapped in selected method.On this basis,the BPSO is used for searching the whole feature space to find the best feature subset.The experiment on DARPA datasets shows the proposed method result approximate to BPSO-SVM and the calculation cost reduced expressly.The result shows ML-BPSO reduce the calculation cost while gets good performance on classification accuracy and dimensional decrease.