基于支持向量机的足月胎儿体重预测新方法  

A NEW METHOD FOR ESTIMATION OF FETAL WEIGHT USING SUPPORT VECTOR MACHINE

作　　者：宋晓峰[1] 韩平[2] 邹丽[2] 陈德钊[3] 胡上序[3] 

机构地区：[1]南京航空航天大学生物医学工程系,南京210016 [2]华中科技大学同济医学院附属协和医院,武汉430022 [3]浙江大学智能信息工程研究所,杭州310027

出　　处：《中国生物医学工程学报》2004年第6期516-522,共7页Chinese Journal of Biomedical Engineering

基　　金：国家自然科学基金资助项目 (2 0 0 760 41)

摘　　要：支持向量机回归估计的性能往往依赖于核函数及其参数、不敏感系数和惩罚因子的确定。支持向量机中参数的确定是一个较为困难的问题 ,以往通常采用交叉验证的方法确定参数。本研究提出了自适应参数调整支持向量回归估计方法 (A -SVM) ,给出了自适应参数调整算法。并已成功地应用于足月胎儿体重的建模 ,且经与径向基网 -偏最小二乘回归方法 (RBFN -PLSR)建立的模型比较 ,A -SVM方法拟合精度和预测能力均比RBFN-PLSR方法好。The performance of support vector machine for regression estimation depends on the kernel function, its parameter, insensitive factor Ε and penalty factor C. The determination of the parameters of SVM remains a difficult problem. Cross - validation method is commonly used in practice to decide the parameters of SVM, however, it is notefficient enough. A novel adaptive support vector machine (A - SVM) was proposed in this work to determine the optimal parameters automatically. The algorithms for adaptively tuning parameters of SVM established. A - SVM was successfully applied in estimating the fetal weight. Compared with RBFN - PLSR methods, A - SVM was superior in both fitt and testing accuracy.

关 键 词：支持向量机 胎儿体重 建模 回归估计 

分 类 号：R319[医药卫生—基础医学]