Shale gas reservoirs are characterized by strong heterogeneity and low porosity and permeability. Hydraulic fracturing is an essential technique for exploration and development of shale gas wells. Accurately predicting the productivity of shale gas wells is a crucial step in evaluating fracturing effectiveness, involving a complex process with numerous influencing factors. To achieve rapid and precise prediction of fracturing productivity, this study proposes a method for predicting the productivity of shale gas wells after fracturing based on ensemble learning. By collecting and organizing a large amount of geological, logging, perforation, fracturing construction, and production data, data cleaning and preprocessing are conducted. Fine parameters describing geological and engineering features are extracted from multiple dimensions and scales. Various correlation analysis methods are employed to analyze the main controlling factors of fracturing productivity in shale gas wells. Based on XGBoost, Random Forest, and Gradient Boosting algorithms, basic models are constructed. And a voting regressor is utilized to aggregate the predictions of multiple basic models, ultimately forming a fracturing productivity prediction model to accurately forecast post-fracturing effects. This research focuses on more than 640 hydraulic fracturing wells in a certain shale gas block. The results indicate that factors such as horizontal section length, organic carbon content, total gas content, fluid intensity, proppant intensity, and number of fracture clusters significantly influence fracturing effects. The accuracy of productivity prediction matches actual productivity by over 90%. The findings of this study can provide a scientific basis for predicting fracturing effects and designing fracturing schemes.