支持向量机法预测卤代苯和苯酚衍生物的毒性

应用支持向量回归算法（SVR）,以按非氢原子分类的分子电性距离矢量（H-MEDV）为参数,通过参数的优化,建立了几种更强的预测模型,预测了卤代苯和苯酚衍生物的毒性,并根据H-MEDV参数原理对预测结果进行了初步探讨.各预测模型的标准误差分别为0.305、0.267、0.275、0.228、0.362、0.238,均低于采用多元线性回归（MLR）和逐步回归（SMR）法的预测结果,说明支持向量回归算法在小样本、多变量的样本建模预报问题上具有一定的优势.

Support Vector Machine Method Applied to Predict the Toxicity of Halogeno Benzene and Phenol Derivatives

Support vector regression algorithm was used to set up some more predictive model through optimizing parameters which is called the hydrogen-association classified molecular electrongativity-distance vector(H-MEDV).The best predicted model was reported for the toxicity of halogeno benzene and phenol derivatives.The root mean suqare error(RMS) of each models were 0.305,0.267,0.275,0.228,0.362,0.238,which were better than that of relective models constructed by multiple linear regression(MLR) and stepwise regression(SMR).The satisfactory results of practical application indicated that the algorithm of SVR were well used to set up predictive model of data mining and multivariable samples.