The melting point (MP) of a chemical is an important physicochemical property that characterizes the transition from a solid to a liquid state. The MP is a key parameter in molecular design and relevant in many fields such as drug design and environmental science. Therefore, an accurate prediction of the MP is of huge interest. Here, we develop two graph convolutional neural network (GNN) models for the prediction of the MP: one where we do not apply a data augmentation strategy and one where we apply a data augmentation strategy. The models were developed on a data set containing 28,645 chemicals, where we removed duplicates and data points labeled as faulty. Then we split the data set into training, validation, and test sets. The model was trained on this initial data set and on a higher curated data set. Based on the data augmentation, we could enlarge the number of neurons in each of the two hidden layers in the GNN and reinforce the representation of large and complex molecules. We compared the influence of the curation step and the data augmentation and found that the curation step had no significant influence on the model performance, while the model could be improved by the application of data augmentation. With a consensus model, we achieved an rmse of 35.4 °C.