Optimizing the Synthesis Parameters of Double Perovskites with Machine Learning Using a Multioutput Regression Model

Double perovskites are attractive candidates for addressing the main challenges associated with efficient single perovskites, such as low stability and the presence of heavy metals in their composition. However, the double perovskite structure is complex and presents unique synthesis challenges. Furthermore, fine-tuning the synthesis parameters to obtain precise control of the nanoparticle size is necessary. To tackle these issues, we combined machine learning with a multioutput model, allowing us to simultaneously generate multiple outcomes within a single regression while considering interactions among all targets in a complex reaction. As a result, we built a specific data set with relevant parameters for the synthesis of double perovskite, using the hot injection method considering lead-free double and single perovskites. We then developed three different multioutput machine learning models based on decision trees, random forests, and neural networks. These models were trained to predict optimal synthesis conditions for double perovskites, including reagent amounts, time of reaction, and bandgap. The selection of the best model was based on MAE, RMSE, and R2 metrics. We utilize this model to predict the synthesis condition of a double perovskite, which we subsequently synthesized, specifically Cs2AgInCl6, hence validating our model with experimental results. Our approach enables us to achieve accurate predictions and gain a deeper understanding of the intricate relationships between synthesis parameters and material properties.