作　　者： ; ; ; ;

机构地区： 重庆邮电大学计算机科学与技术学院

出　　处： 《江苏大学学报（自然科学版）》 2013年第4期410-415,共6页

摘　　要： 针对单分类器识别率低、稳定性差的问题,提出了一种用于微阵列数据分类的演化硬件多分类器选择性集成方法.首先把经过预处理的原始训练集随机划分为训练集和验证集;然后通过对训练集的学习获得基于演化硬件的基分类器;再用验证集评价基分类器的性能,选择其中一部分较好的基分类器组成最终的分类系统;最后用独立的测试集验证系统的性能.试验结果表明,对急性白血病和结肠癌数据集的识别率分别为95.42%、88.33%,与其他的模式识别方法具有可比性;同时在识别率相当的情况下,该方法的硬件代价远低于全集成的演化硬件多分类器. In order to solve the problems of low recognition rate and poor stability in a single classifier, a selective multiple classifiers ensemble-based evolvable hardware was proposed for the classification of microarray data. The proposed original training set was randomly divided into a training set and a validation set. The base evolvable hardware classifiers was trained by the partitioned training sets, and the performance of the trained base classifiers were evaluated by the validation set. The better partial base classifiers were selected to build a final ensembled classifier, and the performance of the ensembled classifier was tested with an independent test set. The experimental results show that the recognition rate of the pro- posed evolvable hardware for the classification of acute leukemia and colon cancer are 95.42% and 88.33% , respectively, which are higher than those of other pattern recognition methods. Compared with traditional multiple classifiers ensemble-based evolvable hardware, the proposed scheme has similar recognition rate, much lower hardware cost.

关 键 词： 模式识别 机器学习 演化计算 分类器 选择性集成 微阵列

领　　域： [自动化与计算机技术] [自动化与计算机技术]