基于RFECV-RF-Boosting的烟叶感官质量预测研究OA

[Objective]This study aimed to address the problems of strong subjectivity and difficulty in data acquisition in sensory evaluation of tobacco leaves,and to achieve precise quantitative prediction of tobacco leaf sensory quality based on chemical composition data.[Methods]A total of 264 tobacco leaf samples from four typical style-producing regions(Henan,Hunan,Yunnan,and Guizhou)were used for chemical composition determination and sensory quality evaluation.After removing redundant indicators through correlation analysis of chemical variables,the recursive feature elimination with cross-validation based on random forest(RFECV-RF)method was employed to select the optimal feature subset for each sensory attribute.Subsequently,three classic boosting algorithms,namely XGBoost,CatBoost,and LightGBM,were applied,and their hyperparameters were optimized via five-fold cross-validation to develop prediction models for nine sensory attributes.[Results]1)RFECV-RF feature selection revealed that total nitrogen,reducing sugars,potassium,and nicotine were the key chemical components influencing tobacco leaf sensory quality.2)Except for"strength,"the RMSE values for all other attributes were lower with the optimal feature subset than with the full feature model.3)Under the optimal algorithm,the coefficients of determination(R2)for the sensory attributes ranged from 0.711 3 to 0.894 0,RMSE from 0.084 5 to 0.140 4,and mean absolute percentage error(MAPE)from 1.06%to 1.70%,all showing good and stable predictive performance.[Conclusion]The proposed prediction model framework enables high-precision quantitative prediction of tobacco leaf sensory quality.These result provide scientifically reliable technical support for digital formulation design and quality control of cigarette products.