Prediction of the Melting Points of Double Halides

Abstract—The atmospheric pressure melting points of binary halides ABHal₃, ABHal₄, A₂BHal₄, A₂BHal₅, and A₃BHal₆ (A and B are various elements; Hal = F, Cl, Br, or I) were predicted. Calculations were made using our developed system of machine learning. Computer models were searched by analyzing information on the already known melting points of halides. The unknown melting points of the halides were predicted using only the values of the properties of the elements A, B, and Hal. It was shown that the programs based on the methodology of ensembles of machine learning algorithms make the most accurate estimates of melting points (the mean absolute errors determined by the cross-validation method in the leave-one-out cross validation mode were in the range of 29–52 K, depending on the halide composition and the chosen algorithm). The coefficient of multiple determination for the models used for prediction was no lower than 0.7.