Aiming at the problems of inaccurate prediction model of finish rolling width spreading due to roller change or other reasons during the rolling process of hot-rolled strip steel, and long training time when the dimension of data collected was high and the amount of data was large in the field of hot rolling, a width prediction method based on incremental learning tree model for operating condition drift was proposed. First, the causes and main characteristic parameters of the variable working conditions were obtained based on the traditional finish rolling mechanism model. Then, the effective data features summarized at the mechanism and data levels were combined as the input of the prediction model, and the width spreading prediction model was established by the incremental learning method. Finally, the tree model established by incremental learning was compared with other machine learning models with the same indicators on actual data by experimental design. The results show that the proposed tree model based on incremental learning has higher prediction accuracy and low running time in wide-spread prediction.