基于粒子群优化算法的视频流特征选择方法

Video traffic feature selection method based on particle swarm optimization

针对网络视频业务流分类存在特征维数过高的问题,提出一种基于ReliefF和粒子群优化算法(PSO)的特征选择方法。该方法首先利用ReliefF算法滤除部分无关特征,达到快速降维的目的,然后以PSO作为搜索算法,并用ReliefF算法得到排名靠前的部分优良特征作为粒子群的部分初始种群,以不一致率作为评价函数在剩余特征子集中选择出最优子集。ReliefF既减小了特征空间维数,又为PSO算法提供了先验知识,从而提高了算法整体的分类精度和搜索效率。实验表明,该方法在不同的数据集上均能保持较优的性能。

To solve the problem of high feature dimensionality of network video traffic, a feature selection method based on ReliefF and particle swarm optimization （PSO） is proposed. Firstly, some irrelevant features are removed to achieve the fast dimension reduction by ReliefF algorithm. Then, the optimal feature is searched on the subset by using some better subsets of ReliefF algorithm as initial PSO population and evaluating feature subsets by inconsistency rate. The ReliefF algorithm can reduce the feature space and provide the prior knowledge of the PSO algorithm, thus improving the search efficiency and the classification accuracy of the algorithm. Experiments show that the method can achieve better performance in different data sets than existing methods.