Abstract: To establish an accurate and widely applicable prediction model for tar and nicotine released from flue-cured tobacco, and analyze the influences of chemical components on the release contents, 455 domestic and foreign flue-cured tobacco samples from four enterprises were selected and rolled into conventional cigarettes. Content or value data of 65 macro and semi-micro components and 5 chemical indexes in tobacco leaves were rapidly predicted from near-infrared spectroscopy, and seven cigarette auxiliary material parameters were collected. The prediction model was established by using multiple machine learning method. The effects of chemical components and indexes were analyzed based on the Shapley Additive ExPlanation (SHAP) method, and the influence of the 70 components and indexes in tobacco leaves on tar and nicotine release contents were clarified. The results showed that: 1) The prediction models for tar and nicotine released from flue-cured tobacco were established using the kernel function-based SVM. The coefficient of determination (R²) of external validation set reached 0.849 and 0.914, with mean absolute errors (MAE) of 0.430 and 0.073, respectively. For blended formula of chemical components linearly added by single material, the relative prediction errors of prediction of all indices were less than 10%. The prediction model obtained by the above methods exhibits excellent wide applicability and high accuracy 2) In the SHAP interpretation, group permutation of highly correlated components can avoid contradictory conclusions. For tar release, the top five influential components and indexes are myristic acid, dichloromethane extract, potassium (K), solanesol, and malic acid. For nicotine release, the top five influential components and indexes are total alkaloids, malic acid, calcium (Ca), myristic acid, and dichloromethane extract. The effects of conventional components are generally consistent with results reported in previous studies.