基于特征分类器集成的糖尿病分类方法

发布时间：2018-03-20 13:10

本文选题：糖尿病　切入点：特征分类　出处：《中华中医药杂志》2016年01期 　论文类型：期刊论文

【摘要】：目的:结合医用电子鼻技术,探讨糖尿病患者及其口腔呼气的气味图谱特征。方法:选择180例糖尿病患者和100例健康者,用医用电子鼻采集280例口腔呼气的气味图谱,采用基于数据特征划分的方法,用支持向量机和随机森林集成模型对糖尿病患者进行分类预测。结果:1线性核函数的支持向量机(SVM1)分类结果不是很理想,低于多项式核(SVM2)、径向基函数核(SVM3)和随机森林(RF)3种分类器,说明分类超平面显然是非线性的;2集成分类器对糖尿病患者和健康者的气味图谱特征的识别准确率可达88.04%。结论:基于特征划分的分类器集成方法预测性能明显好于单一分类器,为使用医用电子鼻进行糖尿病诊断分析提供了一种有效手段。
[Abstract]:Objective: to explore the characteristics of odour atlas of diabetic patients and their oral exhalation with medical electronic nose technique. Methods: 180 diabetic patients and 100 healthy subjects were selected and 280 oral breath breath odors were collected by medical electronic nose. Based on the method of data feature partition, support vector machine (SVM) and stochastic forest integrated model were used to predict the classification of diabetic patients. Results the classification results of the support vector machine SVM1 of the linear kernel function of 1: 1 were not very satisfactory. These classifiers are lower than polynomial kernel SVM2, radial basis function kernel SVM3) and random forest RFN. The results show that the classification hyperplane is obviously nonlinear and the accuracy of recognition of odour map features of diabetes patients and healthy people can reach 88.04.Conclusion: the prediction of classifier ensemble based on feature partition is better than that of single classifier. It provides an effective method for the diagnosis and analysis of diabetes using medical electronic nose.
【作者单位】：福建工程学院信息科学与工程学院;厦门大学自动化系;福建中医药大学中医证研究基地;
【基金】：国家自然科学基金项目(No.81373552) 福建省教育厅A类项目(No.JA14212) 福建工程学院科研启动项目(No.GY-Z12079)~~
【分类号】：R587.1

【参考文献】