Wavelet neural network-based QSPR for prediction of critical micelle concentration of Gemini surfactants Original Research Article

A QSPR study is performed to develop a model, relating the structure of 94 Gemini surfactants to their critical micelle concentrations for the first time. A whole number of descriptors were calculated with Dragon software and a subset of calculated descriptors was selected from 18 classes of Dragon descriptors with a forward stepwise MLR method. The data set was randomly divided into three data sets: calibration (69 molecules), prediction (15 molecules) and test set (10 molecules). The selected descriptors were used as input neurons in a wavelet neural network (WNN) model. The WNN architecture and its parameters were optimized simultaneously. The performance of the model was investigated by the test set and the average absolute error was 0.105 for the test set. The capability of the WNN model was compared with the MLR model. It was demonstrated that the WNN is superior to the MLR model.