Infrared Spectra Interpretation For Conjugation Of Alkenes Based On Feature Selection By Genetic Algorithm-Partial Least Square Approach

Analytical chemistry is the science of studying the method of analysis and getting the chemical component and structural information. Until now from back 70 years, the coming of information era based on the application of computer, which brought great development to science and technology, had resulted in the combination of analytical chemistry and computer. At present, the appearance and abroad application of many modern analytical instrumental make it possible that obtain many information at short notice, which provide opportunity to the extraction of chemometrics. The first application of ANN to the parse of IR was advanced by Munk, other methods such as ANN, PLS, WT, SVM was step by step introduced to the investigation of IR subsequently, which make great development of pattern recognition in IR. GA-PLS was used to the feature selection of alkenes in IR, while SVM and ANN were introduced to the classification with selected feature. In this paper, we design a multiple-level system according to the classification of 899 alkenes from OMNIC database. Firstly, the alkenes were classified to cycloalkenes and chain alkenes; Secondly, the alkenes were further classified in order to identify different conjugate group. Every classification was modeled with full spectrum and selected feature, respectively. The result indicates that GA-PLS was a very good tool of feature selection and the model with selected feature was better than full spectrum, that reduced the complexity of model and increased the forecast ability. We discuss the segmental spectrum in order to further investigate the relationship of the segmental spectrum and structure. After the feature selection by GA-PLS, the feature was selected according to four characteristic spectrum region which were C-H stretch vibration, C=C stretch vibration, C-H wagging in-plane vibration and C-H wagging out-of-plane vibration to assemble for the sake of discuss the segmental spectrum. The result implied that C-H wagging out-of-plane vibration was most important in classifying alkenes are cycloalkenes or chain alkenes. C=C stretch vibration was more vitally in classifying cycloalkenes. In the situation of chain alkenes, C-H stretch vibration was of importance in the classification of chain alkenes conjugate with double-bonds, while C=C stretch vibration was more important in the identification of conjugate and non-conjugate chain alkenes, chain alkenes conjugate with carbonyl, chain alkenes conjugate with benzenes and chain alkenes conjugate with both carbonyl and benzenes.