Application of Radial Basis Function Neural Network in Environmental Chemistry

The retention behaviour of 66 organic pollutants in biocompatible micelles was studied by QSPR method. The linear and nonlinear models between the structures of these compounds and their chromatographic retention values were established by using the heuristic method and the RBFNN method, respectively. The correlation coefficients of the two methods are 0.8400 and 0.8642, respectively, and the corresponding root mean square error (RMS) is 0.1577 and 0.1562. The results show that there is a significant linear relationship between the selected descriptor and the retention value by comparing the stability and prediction ability of the two models. The nonlinear method cannot improve the prediction ability of the model.