Investigation of Factors Controlling Smoking Quality of Flue-cured Tobacco Using CART Models
XU An-ding,DU Guo-wei,LIU Hong-bin
DOI: https://doi.org/10.3969/j.issn.1001-4829.2013.04.012
2013-01-01
Abstract:Smoking quality is the one key quality index for flue-cured tobacco,whereas flue-cured tobacco smoking quality is closely related to chemical component,variety and grade in recent research. In this study,a total of 15 chemical component( total sugars,reducing sugars,nicotine,total nitrogen,protein,potassium,chloride,starch,phosphorus,calcium,magnesium,sulfur,boron,copper and zinc) and 5 item of tobacco smoking quality key index( aroma quality,aroma quantity,taste,impurity and irritation) were determined based on 196 tobacco leaf samples which collected from 30 tobacco counties in Guizhou province,and conventional descriptive statistical analysis,Kolmogorov-Smirnov test,correlation analysis,classification and regression trees( CART) and stepwise regression analysis were performed using SPSS13. 0. The results show that coefficient variance of chemical components varied between 17. 9 % and 90. 5 %,coefficient of variant of smoking quality indicators was less than 6 % in different tobacco samples,chloride,phosphorus and copper followed lognormal distribution and the rest indicators presented normal distribution,the relationships between chemical component and smoking quality indicators varied from significantly positive correlation to significantly negative correlation. According to root square mean error( RMSE),mean absolute error( MAE) and mean absolute percentage error( MAPE),CART model outperformed stepwise regression on forecast precision for flue-cured tobacco smoking quality index in this study. The rank order of the factors affecting tobacco smoking quality was chemical component >variety > grade. The results of this study will provide valuable information for investigating variability of tobacco chemical component,relationships between chemical component and smoking quality,prediction of smoking quality using chemical component and selection of suitable variety at regional scale.