Use of Random forest in the identification of important variables

摘要: Abstract Random Forest (RF) technique has been shown to be promising in the supervised classification applied different matrices. However, approaches identifying significant variables that weight model are scarce, problems. In this paper, we propose a methodology for selection of greater relevance construction RF models. For application methodology, models were developed discriminating crude oil samples, about their maximum pour point (MPP). sense, data from MPP (ASTM D5853) 105 hydrogen (1H) NMR spectra and carbon (13C) acquired. With ranging −54 °C 39 °C, two classes assigned: first containing 43 samples with value ≤ −9 °C, and, second, 62 value > −9 °C. The 1H models, 90% accuracy, 13C NMR, 71% used variable method. results showed proposed select was effective distinction best contributed discrimination oils. Therefore, new tool enabled understanding interest chemical information, contained its relationship property samples.