Computer-assisted prediction of the classification of the pesticide chemical structure in mass spectra

The structural information of pesticides in mass spectra has been investigated. It is of assistance in the analysis of pesticide residues and in the selection of pesticide lead compounds. From GB 4839-1998, four kinds of pesticide structures are chosen containing organochlorine pesticides, organophosphorus pesticides, carbamate pesticides, and pyrethroid pesticides. Mass spectral data selected from NIST 2.0 were transformed into 396 features firstly and then feature selection by Genetic algorithm-partial least squares (GA-PLS) is implemented to get the optimization feature set. At last, prediction models are constructed with K-nearest neighbor method (KNN), Support vector machine (SVM), and AdaBoost algorithm combined with Classification and Regression Tree (AdaBoost-CART) and their advantages and disadvantages are discussed. It is proved that SVM and AdaBoost-CART with the optimization set of few features could get better prediction results.