By making composite blocks of ceramsite lightweight aggregate concrete（LWAC） and normal concrete（NC）， the end-to-end gradient boosting decision tree（GBDT） predicting model was proposed based on the grouping experiment， which correlates composite interface preparing method， casting interval time and normal force on composite interface to bonding strength between LWAC and NC. The results from GBDT bonding strength prediction model are compared with those from the support vector machine regression model， K-nearest neighbor regression model， the decision tree and BP neural network. The comparison shows that the designed GBDT model is more robust than the other four models with superior predictive performance synthetically testified by the goodness of fit， mean absolute error and root mean squared error. In addition， the mean relative error from GBDT prediction for the test samples is obviously smaller than that from the other four models. An effective and satisfactory prediction result can be obtained for the composite interface bonding strength between LWAC and NC from the established GBDT model.