Using Rough Reducts Based SVM Ensemble for SAR of the Ethofenprox Analogous of Pesticide

Abstract

Neural networks ensemble is a promising tool in the field of structure-activity relationship (SAR). Based on support vector machine (SVM), a new method called RRSE (rough reducts based SVM ensemble) is employed to discriminate between high and low activities of ethofenprox analogous based on the molecular descriptors. By using RRSE, individual SVMs of ensemble model are constructed by projection of training dataset on sufficient and necessary attribute sets (reducts). Finally, the results from all individuals are combined by majority voting to finalize the ensemble results which predict activities of ethofenprox analogous with accuracy of 93.5%. Experimental results indicate that performance of RRSE is better than those of SVM bagging, optimal reducts based SVM and single SVM. Therefore, RRSE could be a promising and useful tool in SAR research.