In the process of sintering production，sinter grade is one of the important indicators to evaluate the quality of sinter. In order to improve the level of sintering production，a prediction model of sinter grade proportion based on the overall production parameters is established in order to solve the problems of hysteresis and grade proportion control and prediction in the sintering process. The 12 key parameters after feature selection are taken as input variables，and the corresponding sinter grade proportion is used as the output variable. Through data preprocessing methods such as missing data filling，data augmentation，and outlier point replacement，high-quality datasets for predicting the proportion of sinter fraction are obtained. The Osprey Optimization Algorithm ( OOA) and Categorical Feature Gradient Boosting Algorithm ( CatBoost) are used to construct the prediction model. The results show that the mean absolute error ( MAE) of the OOA-CatBoost algorithm model is 0. 276 9，the mean square error ( MSE) is 0. 043 3，and the coefficient of determination ( R² ) is 0. 949 9. Compared with 11 other machine learning models，such as the dwarf mongoose optimization ( DMO) algorithm，the sparrow search algorithm optimization ( SSA) ，the random forest ( ＲF) optimized by the osprey optimization algorithm，the lightweight gradient booster ( LightGBM) and the extreme gradient boosting algorithm ( XGBoost) ，the proposed model achieves good prediction results. Based on the industrial measured data，the average prediction error of the OOA-CatBoost algorithm for the proportion of sinter grade can reach 0. 085 2，which can provide theoretical guidance for optimizing the ratio of raw materials，mixing parameters and sintering machine parameters，so as to improve the grade proportion of high-quality sinter.