论文部分内容阅读
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直未获根本性解决。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、两描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,且支持向量回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。
Based on the advantages of generalized promotion ability of support vector machines with minimum structural risk, etc., they are widely used in classification and prediction fields. However, their weak interpretations have not been fundamentally solved. A set of complete explanatory system based on F test for support vector regression was established, including model saliency test, single descriptor significance test, single descriptor effect and sensitivity analysis, two descriptor interaction significance test and so on. The quantitative structure relationship of anionic surfactants used in pesticides was studied. The experimental results show that the interpretation results are consistent with the stepwise linear regression model and the quadratic polynomial stepwise regression model, and the performance of the support vector regression model is significantly better than The reference model shows the rationality of the explanatory system.