基于蚁群优化的选择性集成数据流分类方法  

Selective Ensemble Data Stream Classification Based on Ant Colony Optimization

作　　者：王军[1] 刘三民[1] 刘涛[1] 

机构地区：[1]安徽工程大学计算机与信息学院,安徽芜湖241000

出　　处：《长江大学学报（自然科学版）》2017年第5期37-43,共7页Journal of Yangtze University(Natural Science Edition)

基　　金：国家自然科学基金项目(61300170);安徽省自然科学基金项目(1608085MF147);安徽省高校省级优秀人才重点项目(2013SQRL034ZD)

摘　　要：基于集成学习的数据流分类问题已成为当前研究热点之一,而集成学习存在集成规模大、训练时间长、时空复杂度高等不足,为此提出了一种基于蚁群优化的选择性集成数据流分类方法,用蚁群优化算法挑选出优秀的基分类器来构建集成分类模型。该方法首先对所有基分类器采用交叉验证计算分类精度,同时采用Gower相似系数求出基分类器之间的差异性,然后把分类精度和分类器差异性作为分类器挑选标准,从全部基分类器中选出一部分来构建集成模型,最终挑选的基分类器不仅具有良好的分类精度,同时保持一定差异性。在标准仿真数据集上对构建的集成分类模型进行仿真试验,结果表明,该方法与传统集成方法相比在准确率和稳定性方面均有显著提高。Data streams classification based on ensemble learning has become one of the current research hotspots,which exists many disadvantages,such as the large volume,long training time and higher complexity.To solve these problems,the new selective ensemble data stream classification method is presented based on ant colony algorithm.The method uses the classification accuracy and the classifier’s difference as the classifier selection criteria,the ant colony optimization algorithm is used to select the base classifiers with high classification accuracy and large individual difference to construct the ensemble classification model.Lastly,simulation experiment is carried out on the standard dataset,the experiment results show that the presented method is superior to the traditional ensemble methods in classification accuracy and stability.

关 键 词：数据流分类 概念漂移 选择性集成 蚁群优化算法 差异性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]