一种新的选择集成算法  

A New Selective Ensemble Algorithm

作　　者：王道庆 孙浩军[1] WANG Daoqing;SUN Haojun(Department of Computer, College of Engineering, Shantou University, Shantou 515063, Guangdong, China)

机构地区：[1]汕头大学工学院计算机系,广东汕头515063

出　　处：《汕头大学学报（自然科学版）》2019年第1期48-55,共8页Journal of Shantou University：Natural Science Edition

摘　　要：针对一般的选择性集成学习算法在选择分类器阶段需要独立设置验证集因而损失了一定的训练数据的缺点,提出了一种新的选择性集成分类算法FPSE,该算法采用一种基于排序的策略,这种策略在选择阶段就存在速度上的优势;其优势在于不必独立设置验证集,而采取一种将个体选择评估融入在原始数据本身的方法.实验验证了FPSE算法在个体评估策略的有效性,以及较好的泛化性能;对比试验说明了该算法的分类预测表现要优于Bagging算法和AdaBoost算法.For the general selective ensemble learning algorithm, some training data could be losing by the reason of setting the verification set independently in the selection of the classifier. Aiming at the shortage, a new selective ensemble classification algorithm FPSE is proposed. The new algorithm FPSE is a method based on a sorting strategy, which has an advantage of speed in the selection phase. The characteristics of algorithm FPSE is taking a strategy to incorporate selection assessment into the raw data itself, which making it unnecessary to set the verification set independently. The experimental results show that the proposed FPSE algorithm has better generalization performance, and the individual evaluation strategy is efficient. The classification prediction performance of the new algorithm is better than the Bagging algorithm and the AdaBoost algorithm in the comparative experiments.

关 键 词：选择性集成 分类器 泛化性能 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]