<p>It is essential for the molten fluxes to possess appropriate crystallization temperatures to meet the requirements for heat transfer control and lubrication in the mold during continuous casting and electroslag remelting. However, the data measured by various experimental techniques are very limited, and it is unrealistic to obtain a mass of data of crystallization temperatures of a certain slag system in a large composition range. To predict the crystallization temperatures of mold fluxes and electroslag remelting-type slags covering intensive composition ranges, six models for predicting the crystallization temperatures of molten fluxes are established based on six machine learning algorithms (LightGBM, ET, SVM, EN, KNN and CNN), and the prediction accuracy of these six models is compared. The model established based on the LightGBM algorithm shows the highest prediction accuracy, with <i>R</i>², MAE, and RMSE values of 0.969, 16.289, and 24.863, respectively. SHAP feature ranking reveals that the contents of B₂O₃, MgO, Al₂O₃, SiO₂, and Li₂O and cooling rate are the primary factors influencing the crystallization temperatures of molten fluxes. LightGBM model precisely describes the influence of the content of each component in the slag and cooling rate on the crystallization temperatures of the slag melts. A software for predicting the crystallization temperatures of slag melts and designing chemical compositions of fluxes is developed with a comprehensive <i>R</i>² value of 0.991. This software is capable of precisely calculating the corresponding chemical composition of the slag aiming at the target crystallization temperatures and a given cooling rate.</p>