QSPR study of vapor pressure of PBDEs by SVM method

Luo Yuanfeng,Huang Jun,Yu Gang
DOI: https://doi.org/10.16866/j.com.app.chem2009.06.021
2009-01-01
Abstract:With MOPAC-PM3 method in the ChemOffice 8.0 software,6 quantum chemistry parameters for 23 kinds of PBDEs were calculated.Using support vector machine(SVM) method based on polynomial kernel,radial basis kernel,sigmoid kernel,a QSPR model has been set up for vapor pressure of 23 kinds of PBDEs.For the training set,the correlation coefficient R~2 of the three types of kernel function is 0.994,0.996 and 0.994,the mean square error(MSE) is 0.0102,0.0081 and 0.0095,respectively.The correlation coefficient R~2 of Leave One Out(LOO) cross validation is 0.992,0.991 and 0.991,respectively.For the testing set,the correlation coefficient R~2 of the three types of kernel function is 0.994,0.986 and 0.991,the MSE is 0.0225,0.0458 and 0.0247,respectively. The results show that the SVM regression algorithm play an excellent performance on QSPR modeling for the vapor pressure of PBDEs,and the choice of kernel function of SVM model do not significantly affect performance.The overall performance of the three types of kernel functions has been compared through model fitting,prediction ability and the stability,and compared the influence of the support vector number on the prediction ability.The results show that the properties of polynomial kernel and sigmoid kernel are similar,better than radial basis kernel.
What problem does this paper attempt to address?