Effective prestress under the anchorage refers to the prestress force retained by the prestress tendons under the anchorage opening of the working anchorage after the prestress tendons are tensioned and anchored. In order to improve the accuracy of the selection of the effective prestress under anchorage value of post-tensioned method small box girder, based on the measured data of effective prestress under anchorage of small box girder of a highway in Guangdong, four machine learning models, namely, multiple linear regression (MLR), adaptive boosting regression (AdaBoost), random forest (RF) and extreme gradient boosting (XGBoost), are used to comprehensively analyze the factors affecting the value. The results show that the mean square error (MSE) of the extreme gradient boosting model is 0.11, the mean absolute error (MAE) is 0.08, the Standard Deviation of the Residuals (SDR) is 0.33, and the goodness of fit (R-Squared) is 0.99. It means that the model predicts accurately, so it can be applied to the problem of accurately fetching the value of the effective prestress force under the anchorages of post-tensioned small box girders under certain conditions. In the XGBoost model, the degree of contribution of each factor to the model value is evaluated by introducing the Gini impurity (Gini) value, and the results show that the thickness of end web is the most important factor affecting the value of effective prestress under anchorages, accounting for 21.65%.