GBDT the tree is regression tree (not the classification tree), GBDT used for regression prediction, after adjustment can also be used for classification. When squared error loss function, each a regression tree learning the conclusions and before all residual trees, obtained by fitting a regression tree residual current, the residual significance Equation: residual = true value - forecast value.
https://zhuanlan.zhihu.com/p/30654833
Interview Highlights of machine learning algorithms GBDT Summary - Part I
The difference between classification and regression tree tree