Feature Selection and Activity Prediction in Chinese Medicine Research Using a Hybrid Model GA-SVM

A new hybrid method called GA-SVM was proposed which combines GA (Genetic Algorithm) as a feature selection model and SVM (Support Vector Machine) as a regression model. With some modifications to the general GA and SVM models, this method can implement feature selection and activity prediction simultaneously, and its performance can be improved. Two experiments are carried out which indicate its better performance than traditional models such as BP net in small sample sets. Important features could be selected by GA-SVM which are validated by MLR. By applying the new model in Chinese medicine research, the QSAR of COX-2 and PGE2 inhibitors were found out and useful conclusions to instruct practical pharmaceutics were drawn.