Precise control of bainitic content and morphology is important for the development of high strength steels with good combination of strength, toughness, and ductility. However, due to elusive mechanisms controlling the bainitic morphology, classical theories of physical metallurgy are inadequate to provide a complete prediction of the bainitic transformation. Here, we propose a machine learning approach to predict both fractions of phase transformation products and their morphologies for different compositions and process parameters after hot-rolling of steel. The modelling strategy is firstly to transform reheating and deformation parameters into physical factors such as parent austenite grain size and stored deformation energy, which can be more directly related to the phase transformation behavior. Secondly, a stacked machine learning model to classify phase transformation products and predict their components under different rolling and cooling conditions was developed using gradient boosting tree classification and support vector machine algorithms. The model is capable of discriminating the type of bainite and the phase fractions for different compositions and process parameters. The model is lastly applied to designing a few lean steel alloys and to optimizing their processing routes, which is validated through analyses of the final microstructure and mechanical properties.