Comparison of Genetic Algorithm Based Support Vector Machine and Genetic Algorithm Based RBF Neural Network in Quantitative Structure-Property Relationship Models on Aqueous Solubility of Polycyclic Aromatic Hydrocarbons

Abstract A modified method to develop quantitative structure-property relationship (QSPR) models of organic contaminants was proposed based on genetic algorithm (GA) and support vector machine (SVM). GA was used to perform the variable selection and SVM was used to construct QSPR model. In this study, GA-SVM was applied to develop the QSPR model for aqueous solubility ( S w, mg•l-1) of polycyclic aromatic hydrocarbons (PAHs). The R 2 (0.980), SSE (2.84), and RMSE (0.25) values of the model developed by GA-SVM indicated a good predictive capability for logSw values of PAHs. Based on leave-one-out cross validation, the results of GA-SVM were compared with those of genetic algorithm-radial based function neural network (GA-RBFNN). The comparison showed that the R 2 (0.923) and RMSE (0.485) values of GA-SVM were higher and lower, respectively, which illustrated GA-SVM was more suitable to develop QSPR model for the logSw values of PAHs than GA-RBFNN.