A steel pipe roundness prediction method according to the present invention includes a step for predicting the roundness of a steel pipe after a pipe expansion step using a roundness prediction model trained using machine learning, including one or two or more parameters selected from attribute information of a steel plate, one or two or more parameters selected from operation parameters of a press-bend step, and one or two or more parameters selected from operation parameters of a seam gap reduction step, as input data, and using roundness information of the steel pipe after the pipe expansion step as output data.