Using the melting point data of dozens of organometalis as a training set, the molecular structure descriptors are extracted according to their SMILES, and the QSPR model is constructed using decision tree-based ensemble learning methods. The model shows a prediction error (MAE) of 33.9 ° C on the test set.