摘要
为提高复杂数据分类器的分类性能,结合人工免疫系统(Artificial immune system,AIS)的自适应识别能力与全局搜索能力,以及近邻传播(Affinity propagation,AP)算法自动确定最佳数据类数的能力,提出了一种基于人工免疫系统与近邻传播相结合的分类算法。通过自适应免疫算法,获得反映数据集模式特征的抗体记忆集,然后再利用基于聚类有效性指标的AP算法确定抗体记忆集的最佳聚类数,以此构造分类器。最后,通过人工数据集和UCI基准数据集来测试该分类器。实验结果表明,与直接采用免疫算法和AP算法相比,该算法在分类正确率和识别性能方面均有良好的表现。在与一些经典分类算法的对比实验中,本文算法也表现出较好的竞争力。
To improve the classification performance of complex data classifier, a classification algorithm based on the combination of artificial immune system （AIS） and affinity propagation（AP） algorithm is proposed which combines the global search capability and adaptive recognition of AIS with the capability of automatically determining the optimal number of data classes of AP. The antibody memory set reflecting the characteristics of data set is obtained by adaptive immune algorithm. Then the optimal number of clusters of the antibody memory set is obtained using AP algorithm based on cluster validity index, and a classifier is constructed. Finally, the proposed algorithm is tested on artificial data set and UCI data set. Compared with artificial immune algorithm and AP algorithm, it has a great competitiveness on recognition rate and recognition capability. The algorithm also shows good competitiveness compared with some of the classic classification algorithms.
出处
《南京航空航天大学学报》
EI
CAS
CSCD
北大核心
2013年第2期232-238,共7页
Journal of Nanjing University of Aeronautics & Astronautics
基金
国家自然科学基金(11078001)资助项目
国家高技术研究发展计划("八六三"计划)(2012AA121602)资助项目
关键词
人工免疫系统
近邻传播算法
K近邻算法
分类
artificial immune system
affinity propagation algorithm
K-nearest neighbor algorithm
classification
作者简介
通信作者：储岳中，男，博士，副教授，1971年生，E—mail：mychu@126．com。
