In order to ensure national food security and promote sustainable agricultural development， it is crucial to accurately predict regional rice yield by formulating effective food strategies. While advancemnets have been achieved by integrating remote sensing data with machine learning technology for yield prediction， a more in-depth analysis of the machine learning model mechanism and the application of high spatio-temporal resolution data remains necessary. This study utilized Sentinel and MODIS Normalized Difference Vegetation Index （NDVI） data along with county-level yield statistics from 2023 to implement a Random Forest （RF） model for predicting yield from county level to pixel level with case study of Sheyang unmanned farm in Jiangsu Province， and investigated the impact of different features on the model's learning mechanism. The results demonstrated that there was a significant correlation between NDVI data and rice yield during the period of August to October 2023， which is a critical period in the crop phenological cycle. The RF model effectively predicted rice yield at the county level， with a Root Mean Square Error （RMSE） value of 339.5 kg /hm². Furthermore， spatial distribution mapping of the predicted yields within Sheyang County for 2023 revealed significant heterogeneity， indicating lower yields in marginal regions and higher yields in central areas ranging from 9 000 to 9 300 kg/hm². This study not only enhances understanding of machine learning application and vegetation data in yield prediction but also provides theoretical support for improving model accuracy and formulating scientific food production strategies.