Abstract: To improve the accuracy of functional classification of tobacco lamina raw materials, an auxiliary classification method based on multisource quality features and ensemble learning was developed. A total of 299 tobacco lamina samples from a cigarette manufacturer were analyzed for chemical composition and sensory quality. Lasso and Boruta were used for dimensionality reduction and selection of model input features. Five classifiers—logistic regression (LR), support vector machine (SVM), random forest (RF), extremely randomized trees (ET), and extreme gradient boosting (XGBoost)—were compared using repeated stratified five-fold cross-validation. A Soft Voting ensemble model was further developed. The results showed that: (1) Thirteen features jointly selected by Lasso and Boruta were retained, comprising eight chemical composition variables and five sensory quality variables. (2) The Soft Voting model achieved the best overall classification performance among the evaluated models. The areas under the receiver operating characteristic curves for quality and aroma enhancement (Class A), smoke balance (Class B), and enhancement of smoke concentration and diffusiveness (Class D) were 0.976, 0.955, and 0.962, respectively. (3) SHAP analysis showed that the key discriminative features and the directions of their contributions differed among functional classes. Higher values of aroma quality, aroma intensity, and fineness contributed positively to Class A predictions; lower values of strength and aroma intensity, together with higher values of succinic acid, off-note score, and cryptochlorogenic acid, contributed positively to Class B predictions; and higher values of strength and myristic acid, together with lower values of cryptochlorogenic acid, contributed positively to Class D predictions. (4) The validation results showed that the agreement rates for Class A, Class B, and Class D samples were 100.00%, 86.67%, and 85.71%, respectively, with an overall agreement rate of 90.00% for the model predictions. This study can provide a reference for leaf-blend formulation and the targeted use of raw materials.