期刊论文

中文

农业科学与技术(英文版)

1009-4229

2016

17

11

2452-2456

10.16175/j.cnki.1009-4229.2016.11.005

Supported by Industrialization Cultivation Projects in Colleges and Universities of Hunan Province（13CY030）； Natural Science Foundation of Hunan Province（12JJ6026）； Colleges and Universities Open Innovation Platform Fund of Hunan Province（14K053,15K066）~~；

本校为其他机构

基于结构风险最小的支持向量机具泛化推广能力优异等诸多优点，在分类和预测领域应用广泛，但其可解释性差的缺陷一甑末获根本性解决。基于F测验为支持向量回归建立了一套完整的解释性体系,包括模型显著性测验、单描述符重要性显著性测验、单描述符效应及灵敏度分析、两捕述符互作显著性测验等。将其应用于农药中常用阴离子表面活性剂的定量构质关系研究，实例验证表明,其解释结果与逐步线性回归模型及二次多项式逐步回归模型结果基本一致，且支持向量同归模型性能明显优于参比模型,表明了该解释性体系的合理性。

In order to solve the problem of poor interpretability of support vector re- gression （SVR） applied in quantitative structure-property relationship （QSPR）, a com- plete set of explanatory system for SVR was established based on F-test, The nov- el explanatory system includes significance tests of model and single-descriptor im- portance, single-descriptor effect and sensitivity analysis, and significance tests of interaction between two descriptors, etc. The results of example indicated that the explanatory results of the new system were co...