Perovskite materials have exhibited promising potential in various fields, including solar cells, photodetectors, and light-emitting devices, owing to their exceptional optoelectronic properties and low-cost synthesis methods. In this work, the quest for thermodynamic stability in perovskite materials for optoelectronic applications is addressed through a machine learning (ML) approach. We also analyzed the importance of cations in the stability of double perovskites. It was found that the AUC (area under the curve) and accuracy of the LightGBM model in classification are 0.918 and 0.92, respectively, and the RMSE (root mean square error) and R2-score in regression are 0.119 and 0.872, respectively. Subsequently, the model was used to predict the properties of 10 new perovskites with high accuracy. Finally, the analysis of cationic sites revealed that the stability of the double perovskite is significantly affected by the elements occupying the A, B, and B’ sites. Our results provide an optimal machine learning approach for discerning the thermodynamic stability of double perovskites, which can be a useful guide for tuning the properties of perovskite materials.