基于自适应粒子群算法的特征选择研究

发布时间：2019-06-22 10:44

【摘要】：在模式分类问题中,数据往往存在许多不相关或是冗余的特征,从而影响分类的准确性。特征选择作为解决这一问题的有效手段,一直以来都是机器学习中的热点。随着数据规模的增加,原始的特征选择方法已经不满足要求。特征选择可以视为一个动态寻优的过程,而粒子群优化算法是目前群体智能算法中的一个热门的算法,由于其简单、易实现、寻优效率高等特点受到了广泛的关注。粒子群优化算法与特征选择方法的结合也成为了一个研究热点。大量的研究表明了基于粒子群优化算法与特征选择结合是可行的,并且有着良好的性能表现。本文主要在粒子群优化算法本身的改进和特征选择问题与粒子群优化方法的结合两个方面做了一定的工作。首先是对粒子群算法的改进,普通的粒子群算法由于其局限性,往往容易陷入局部最优,在骨干粒子群算法的基础上,提出一种基于干扰因子的自适应粒子群算法,在算法的初始过程中引入混沌模型增加初始粒子的多样性,同时在更新机制中引入自适应因子增加其全局搜索能力,提高算法的寻优效率。其次改进粒子群算法中粒子的局部和全局最优的迭代公式。在更新过程中引入对于特征数目的讨论,特别是在解码过程中引入互信息筛选特征,精简特征子集。特征选择的目的在于利用最少的特征达到最佳的优化效果,以往的研究过程中,因为追求更好的分类效果,而忽视特征子集中的特征数目。最后提出一种基于混合模式评价机制的特征选择算法。将特征选择过程分为两个阶段,第一阶段采用基于粗糙集的过滤模式评价机制,第二阶段采用基于邻近算法的封装模式评价机制。为了验证上述提出的理论,选择不同类型的数据集上进行分类实验,得到的实验结果验证了所提出算法的有效性和实用性。
[Abstract]:In the problem of pattern classification, there are often many unrelated or redundant features in the data, which affects the accuracy of classification. As an effective means to solve this problem, feature selection has always been a hot spot in machine learning. With the increase of data scale, the original feature selection method no longer meets the requirements. Feature selection can be regarded as a dynamic optimization process, and particle swarm optimization algorithm is a hot algorithm in swarm intelligence algorithm at present. because of its simplicity, easy implementation and high efficiency, particle swarm optimization algorithm has attracted extensive attention. The combination of particle swarm optimization algorithm and feature selection method has also become a research focus. A large number of studies have shown that the combination of particle swarm optimization algorithm and feature selection is feasible and has good performance. In this paper, some work has been done on the improvement of particle swarm optimization algorithm itself and the combination of feature selection problem and particle swarm optimization method. The first is to improve the particle swarm optimization algorithm. Because of its limitations, the ordinary particle swarm optimization algorithm is often easy to fall into local optimization. On the basis of the backbone particle swarm optimization algorithm, an adaptive particle swarm optimization algorithm based on interference factor is proposed. In the initial process of the algorithm, chaos model is introduced to increase the diversity of the initial particles, and at the same time, the adaptive factor is introduced into the update mechanism to increase its global search ability. Improve the optimization efficiency of the algorithm. Secondly, the local and global optimal iterative formulas of particles in particle swarm optimization are improved. In the process of updating, the discussion of the number of features is introduced, especially the mutual information filtering features are introduced in the decoding process to simplify the feature subset. The purpose of feature selection is to achieve the best optimization effect by using the least features. In the previous research process, the number of features in the feature subset was ignored because of the pursuit of better classification effect. Finally, a feature selection algorithm based on hybrid pattern evaluation mechanism is proposed. The feature selection process is divided into two stages. In the first stage, the filtering mode evaluation mechanism based on rough set is adopted, and in the second stage, the encapsulation mode evaluation mechanism based on neighborhood algorithm is adopted. In order to verify the above theory, different types of data sets are selected for classification experiments, and the experimental results verify the effectiveness and practicability of the proposed algorithm.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】