Toward a Quantitative Model and Prediction of the Cloud Point of Normal Nonionic Surfactants and Novel Gemini Surfactants with Heuristic Method and Gaussian Process

Abstract

A systematic study of integrating statistical modeling and experimental analysis to investigate the cloud point (CP) and environmental risk of 82 structurally diverse nonionic surfactants is performed. During this procedure, the structural profiles of the studied compounds are characterized using hydrophilic domain and the whole molecular. Hundreds of descriptors, including constitutional, topological, geometrical, and electrostatic were calculated by the CODESSA program, and the resulting variables of the characterization selected by heuristic method are then modeled by the Gaussian process (GP). A variety of regression techniques, including MLR, PLS, SVM, and LSSVM are performed to a comprehensive comparison with GP on the basis of statistical analysis and experimental properties, in conjunction with the sophisticated variable selection methods, that is, empirical heuristic strategy. Among all the built models, the most predictable one is constructed based on the GP modeling combination of heuristic variable selection related to hydrophilic domain, with its predictive coefficient of determination () and root-mean-square error of prediction (RMSP) on external independent test set of 0.962 and 5.200, respectively. The statistic model shows that the CP phenomenon is a comprehensive interaction of relative molecular weight, moments of inertia A, and topological structure account for hydrophilic part.