The accurate non-linear quantitive structure-property relationship model for predicting the adsorption constant of volatile and semivolatile organic vapors in soil was firstly developed based on support vector machine (SVM) by using the compounds' molecular descriptors calculated from the structure alone and the features of soil and air. Multiple linear regression (MLR) was used to build the linear QSPR model. Both the linear and non-linear models can give satisfactory prediction results: the correlation coefficient R was 0.953 and 0.995, the mean square error (MSE) was 0.0517 and 0.0057, respectively, for the whole dataset. The prediction result of the SVM model was better than that obtained by the MLR model, which proved non-linear model can simulate the relationship between the structural descriptors, the environmental condition and the soil/air distribution more accurately as well as SVM was a useful tool in the prediction of the adsorption constant of compounds.
- Adsorption constant
- Support vector machine