The CSR and CRI performance index of coke play a crucial role in improving blast furnace smelting efficiency， reducing production costs and promoting green production. Traditional reliance on manual experience regulation has problems such as slow response and large errors，making it difficult to achieve accurate real-time prediction for CSR and CRI. Based on the basic properties of coke，such as ash，volatile content，sulfur content and fixed carbon，big data fitting and machine learning technology are used to construct a decision tree algorithm for predicting CSR and CRI. Grid search combined with cross-validation is used to optimize hyperparameters， the optimal decision tree is screened， and the prediction mechanism of the model is explained by using feature correlation heat map， feature correlation and SHAP value. The results show that when the tree depth is 19 and the number of random seeds is 44，the prediction effect of the CSR model is the best，with an accuracy of 98. 543%. When the tree depth is 18 and the number of random seeds is 75，the prediction effect of the CSR model is the best，with an accuracy of 96. 825% . The field test results show that the single prediction time of the online real-time prediction software after packaging is only 0. 1 ~ 0. 3 s，and the software has good real-time performance and stability. The prediction system developed in this paper can effectively support the quality decision of coke in blast furnace production，significantly improve the prediction efficiency and accuracy，and promote the development of the ironmaking process to be intelligent，green and efficient.