In the financial business, effective risk avoidance and mitigation are perennial themes. In the face of the international economic and financial environment has become more complex and severe, extreme events caused by the tail risk brought about by the harm is very strong, China's banking industry risk prevention and control work will face new challenges. Therefore, under the current globalization and complex and severe economic and financial environment, it is of great practical significance to improve the prediction ability of yield risks of commercial banks and take timely measures to prevent and resolve the risks. Selecting the monthly data of China's A-share listed commercial banks from January 2013 to December 2022, a nonparametric Expectile Regression Forest (ERF) risk prediction model based on the Bagging algorithm to measure the tail risk of China's commercial banks, and simultaneously incorporates the bank leverage ratio, asset size, economic policy uncertainty, financial market volatility and liquidity into the model at the same time were constructed, and then the mean absolute error and root mean square error of the training set and test set data under different models and risk levels respectively was calculateed. Finally, the results were compared with the traditional expected quantile regression (ER) model and Expectile Regression Tree (ERT) model to determine the respective predictive performance. The results show that the ERF model has outstanding performance in measuring the tail risk of commercial banks, and the estimation and prediction ability of the ERF model is significantly better than that of the ERT and ER models under different levels of risk. Further analysis reveals that the smallest and the largest errors in the prediction of tail risk of the four major types of commercial banks are those of the national large-scale commercial banks and the local rural commercial banks, respectively, and the error values of the joint-stock commercial banks and the local urban commercial banks are comparable. The error values of joint-stock commercial banks and local urban commercial banks are comparable.