Abstract: After the Fukushima nuclear accident in Japan in 2011, in order to make up for the inherent defects of UO2-Zr fuel, a new fuel has been developed to improve the capability of reactor core to withstand serious accidents. The development and research of accident tolerant fuel (ATF) has received more and more attention. SiC ceramic microencapsulated fuel pellets show unique advantages in the field of ATF, and is widely studied as a fuel core due to its high thermal conductivity, low density variation with temperature, and good specific heat properties. However, the development of new nuclear fuels has been hampered by the high cost and time-consuming of nuclear fuel experiments. Recently, data-driven approaches can greatly improve the efficiency of new material design and development, such as material property prediction, processing parameter optimization, descriptor extraction and materials design. In this paper, an interpretable machine learning framework was developed to predict sintering performance and optimize process parameters. First, sintering process parameters and sintering performance data were collected from the experiments. Then, based on dataset, a machine learning model between process parameters and the sintering performance of SiC-ceramic microencapsulated fuel particles was constructed. After that, well-trained model was used to predict the sintering performance under different process parameter conditions. Specifically, in order to make machine learning models interpretable and trustfull, feature importance analysis was developed to compute the importance of each process parameter, elucidating the important factors affecting the sintering performance. The results show that the goodness of fit between the predicted and experimental values of the machine learning model on the test set is R²>0.99, RMSE and MAE are about 0.01, and the parameters affecting the sintering performance obtained from the analysis are consistent with the physical knowledge, indicating the accuracy and interpretability of the proposed framework. In addition, a particle swarm optimization (PSO) algorithm was used to search the process parameter space and obtain the process parameters that should have the optimal sintering performance. Lastly, in order to verify the effectiveness of inverse design, pellet density and metallographic diagram under the recommended optimum process parameters were obtained through experimental characterization. The experimental results show that the recommended process parameters with PSO are consistent with experimental results and achieving good sintering performance, demonstrating the feasibility and accuracy of the propsed framework. This work provides theoretical guidance for accelerating the development of new nuclear fuel and also paves a way for the design and development of other materials.