基于Stacking策略的稳定性分类器组合模型研究

Research on Stability Classifier Combination Algorithm Based on Stacking Strategy

针对Bagging、AdaBoost等通用的集成算法对于稳定性分类算法集成效果不是很好的问题,提出了基于Stacking策略的稳定性分类器组合算法.该算法通过构造一个两层的叠加式框架结构,融合数据降维技术处理两层分类器的输入特征,对4种稳定性分类器(LDA、GLM、SVM、KNN)进行组合学习.利用UCI数据集测试算法的性能.实验结果表明:相比一些集成算法(RF、Bagging、C50、AdaBoost),基于Stacking策略稳定性分类器组合模型可以获得更高的分类准确率.同时也为二分类的分类模型提供了一个可行的参考方法.

Aiming at the problems that the general integration algorithms such as Bagging and AdaBoost are not very good for the stability of the classification algorithm,this paper presents a research on the stability classification algorithm based on Stacking strategy.This algorithm constructs a two-layer superposition frame structure,and combines the data dimension reduction technique to deal with the input characteristics of two-level classifier.The four kinds of stability classifiers(LDA,GLM,SVM,KNN)are studied in combination.UCI data set to test the performance of the algorithm.Experimental results show that compared with some integrated algorithms(RF,Bagging,C50),the classification accuracy of the stability classifier based on the Stacking strategy can be improved.It also provides a feasible reference method for the dichotomous classification model.