In the context of low-carbon development, the study of regional carbon emission prediction models is of great significance in guiding the formulation and implementation of future dual carbon target tasks. The ElasticNet- XGBRegressor model, which combines the ElasticNet model as a feature selection model and the XGBRegressor model for regional carbon emission prediction, is a type of ensemble learning model. Based on the principles of the STIRPAT model and the IPCC emission factor method, an original dataset containing 25 features is constructed for the study of regional carbon emission prediction. To validate the effectiveness of the proposed model, an empirical controlled experiment was conducted,with the ElasticNet- XGBRegressor model as the experimental group, and Spearman feature selection and common machine learning methods as the control group. The results showed that the ElasticNet- XGBRegressor model out performed the control group in terms of model evaluation metrics such as RMSE, MAPE, and R2,demonstrating the superiority of the ElasticNet- XGBRegressor model in regional carbon emission prediction. Regression models are innovatively combined with decision tree-based ensemble learning models, leveraging the feature selection capability of the ElasticNet model and the high accuracy and robustness of ensemble learning to improve the accuracy and stability of the prediction model.