基于多目标微粒群优化的异质数据特征选择

环境和测量仪器精度的影响，使得采样数据的不同特征具有不同的质量.对这类异质数据进行特征选择，需要同时考虑特征子集确定分类器的准确度和可靠性，从而增加了特征选择的难度.本文研究异质数据的特征选择问题，提出一种基于多目标微粒群优化的特征选择方法.该方法首先以特征选择的概率为决策变量，将具有离散变量的特征选择问题，转化为连续变量多目标优化问题；然后，采用微粒群优化求解时，基于高斯采样，产生微粒的全局引导者，以提高Pareto解集的分布性；最后，依据储备集中元素更新的速度，确定需要扰动的微粒，以帮助微粒群跳出局部最优.将所提方法应用于多个典型数据集分类问题，实验结果表明了所提方法的有效性.

Feature Selection of Heterogeneous Data Based on Multi-Objective Particle Swarm Optimization

Different features of a sampling datum have different quality as a result the influence of the environment and the equipment precision.For the feature selection of this kind of heterogeneous data,both the accuracy and the reliability of the classifier determined by a feature subset are required to simultaneously consider,which enhances the difficulty of selecting features.The problem of the feature selection of heterogeneous data is focused on in this paper,and a method of selecting features is presented based on multi-objective particle swarm optimization.In this method,the above problem is first converted to a multi-objective optimization problem by regarding the probability of selecting a feature as the decision variable.When particle swarm optimization (PSO) is employed to solve the converted problem,the global guider of particles is generated by Gaussian sampling so as to improve the performance of Pareto solutions in distribution.In addition,the particle to be disturbed is determined according to the speed of updating a particle in the archive to help the swarm jump out of local optima.The proposed method is applied to classify several benchmark data sets,and the experimental results demonstrate its effectiveness.