In steel industry different alloys and heat treatments are used to create materials which fulfil certain properties. A combination of high strength and good ductility is a desired property and can be found in carbide-free bainitic (CFB) steel. The heat treatment allowing the microstructure of CFB steels to form can take up to days, which is not suitable for industry use. Therefore the transformation time expressed in the Time-Temperature-Transformation (TTT) is of great importance to analyse the behaviour of steel alloys during heat treatment. However, the determination of the TTT is expensive and takes a lot of time. This process requires precise measurement of the dilatation of specimens at various temperatures. Past research has used Machine Learning (ML) approaches and physical models to predict the behaviour of the whole TTT. This is mostly done by dividing the TTT into smaller subproblems. These subproblems result from the inherent complexity of the TTT and cover microstructures like martensite, ferrite and bainite. We specifically target predicting TTT for CFB steels, by using a dataset of 56 different material compositions. Random Forests (RFs) and Gaussian Process Regressors (GPRs) are used to predict transformation times. GPRs provide a measure for uncertainty additional to its predictions, which can give insights about how confident the ML model predicts the transformation times. The RF is used because of fast training behaviour and simple hyperparameter tuning. Furthermore, both models allow insight into the significance of specific input features. The feature importance from the RF or the trained length-scales from the GPR are used to determine which input features are important for the model's predictions. The findings of the thesis indicate that, for most predictions, the models are sufficiently accurate. However, the sparsity of the data prevents the model from being precise in certain areas of the feature input space, particularly for certain alloys. Both the GPR and the RF approach have significant potential to allow finding suitable CFB steels with short transformation times. Further research and the creation of a more comprehensive database of measurements will enhance the prediction performance.