摘要
目的研究用于处理解释变量与反应变量之间非线性关系或复杂关系的一种基于核函数的回归方法:核偏最小二乘回归。方法运用Monte-Carlo模拟方法,对核偏最小二乘回归的模型拟合效果和预测效果予以分析。结果模拟试验结果表明:核偏最小二乘回归估计性能均较高。结论核偏最小二乘回归是基于核函数的非线性回归方法,模型构建基于样本,而非解释变量空间,该方法特别适合于处理医学研究中各种类型资料,能够有效地处理解释变量与反应变量之间的非线性关系或复杂关系等方面。
To study kernel partial least squares regression （KPLSR）, a method based on kernel function to deal with nonlinear or complicated relationship between predictor and response variables. Methods The Monte-Carlo method was used to evaluate the effect of KPLSR on the goodness-of-fit and model prediction. Results The simulation results indicated that KPLSR had efficient estimate. Conclusion KPLSR is a nonlinear regression method based on kernel function. Model is built upon samples, not predictors. Especially, this method is suitable in a lot of research fields, such as analyzing all kinds of data in medical research, modeling nonlinear or complicated relationship between predictor and response variables and so on.
出处
《中国卫生统计》
CSCD
北大核心
2007年第3期239-242,共4页
Chinese Journal of Health Statistics
基金
国家自然科基金项目(项目编号:30271157)
关键词
核偏最小二乘回归
核函数
非线性
Kernel partial least squares regression
Kernelfunction
Nonlinearity
