基于人工蜂群算法的支持向量机集成研究

发布时间：2018-02-27 16:47

本文关键词： 人工蜂群算法特征选择支持向量机同步优化集成学习　出处：《湖北工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：支持向量机(Support Vector Machine,SVM)是一种建立在统计学习理论基础上适用于小样本情况的机器学习技术,已经被广泛地应用于模式识别各领域。SVM分类器的性能很大程度上受其自身参数和使用特征的影响,传统方法是将参数寻优问题和特征选择问题进行分开处理,难以得到分类性能整体最优的SVM,但是随着优化计算技术在模式识别领域中的广泛应用,将参数寻优和特征选择问题进行同步优化已变成了一种趋势。另外,由于实际问题的复杂性,SVM的泛化能力也需要进一步提高。集成学习为提高分类系统的泛化能力提供了一条新途径,它通过训练和组合多个有差异的分类器,从而提高分类器的性能,已经取得了较好的进展和成果,然而相关工作并未完善,值得进一步研究。从这一现状出发,本文主要研究了使用人工蜂群算法对支持向量机进行参数特征同步优化和集成研究。首先研究了使用人工蜂群算法(Artificial Bee Colony Algorithm,ABC)进行特征选择和支持向量机参数优化。进而将SVM的参数寻优问题和特征选择问题视为最优化问题同步处理,在提高SVM分类精度的同时尽可能选择少的特征数目,获得整体性能最优的SVM参数和特征子集。为了进一步的提高SVM分类系统的泛化能力,在实现特征参数同步优化的基础上,再引进加权投票集成学习技术,分别构建若干个SVM分类器,在对每个SVM分类器进行学习后,得到若干个具有差异性的SVM分类器,并设置单个SVM分类器的集成投票权重为每个SVM分类器的分类准确率和总分类器数目的比值,将若干个具有差异性的SVM分类器采用加权投票规则的方式进行组合,以期能够得到更优的集成分类性能。为了验证所提出方法的性能,利用部分UCI数据集进行实验验证,本文还将ABC算法与常用的遗传算法和粒子群优化算法进行了对比分析。实验研究结果显示,将其与遗传算法和粒子群优化算法相比,ABC算法在SVM分类器的优化中具有更好的表现;进一步,基于ABC-SVM的加权投票集成算法具有很好的自适应性和分类精度,能够提高基本SVM分类器性能的同时选择出更少的特征数目,并获取整体性能最优的SVM参数和特征子集。
[Abstract]:Support Vector Machine (SVM) is a machine learning technology based on statistical learning theory. The performance of SVM classifier has been widely used in various fields of pattern recognition. The performance of SVM classifier is greatly affected by its own parameters and usage features. The traditional method is to deal with the problem of parameter optimization and feature selection separately. It is difficult to obtain the SVM with overall optimal classification performance, but with the wide application of optimization computing technology in the field of pattern recognition, it has become a trend to synchronize parameter optimization and feature selection. Because of the complexity of practical problems, the generalization ability of SVM also needs to be further improved. Integrated learning provides a new way to improve the generalization ability of classification systems. In order to improve the performance of the classifier, good progress and achievements have been achieved, but the relevant work is not perfect, which is worthy of further study. This paper mainly studies the parameter synchronization optimization and ensemble research of support vector machine using artificial bee colony algorithm. Firstly, we study the feature selection and parameter optimization of support vector machine using artificial Bee Colony algorithm. Furthermore, the parameter optimization problem and the feature selection problem of SVM are regarded as the synchronization processing of the optimization problem. In order to improve the generalization ability of SVM classification system, we can improve the accuracy of SVM classification and select as few feature numbers as possible, and obtain the best SVM parameters and feature subsets. In order to improve the generalization ability of SVM classification system, we can realize the synchronization optimization of feature parameters. Then the weighted voting ensemble learning technique is introduced to construct several SVM classifiers respectively. After learning each SVM classifier, several SVM classifiers with differences are obtained. The integrated voting weight of a single SVM classifier is set as the ratio of the classification accuracy of each SVM classifier to the number of general classifiers, and several SVM classifiers with differences are combined by weighted voting rules. In order to verify the performance of the proposed method, some UCI data sets are used to verify the performance of the proposed method. The ABC algorithm is compared with the usual genetic algorithm and particle swarm optimization algorithm. The experimental results show that compared with the genetic algorithm and particle swarm optimization algorithm, the ABC algorithm has better performance in the optimization of SVM classifier. Furthermore, the weighted voting ensemble algorithm based on ABC-SVM has good adaptability and classification accuracy. It can improve the performance of the basic SVM classifier and select fewer features, and obtain the best SVM parameters and feature subsets.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP18

【参考文献】