High-Accuracy Machine Learning Projections of Composition-Dependent Thermal Stability in Halide Perovskites

Halide perovskites exhibit unpredictable properties in response to environmental stressors due to several composition-dependent degradation mechanisms. In this work, we combine high-throughput experiments, data visualization, and machine learning (ML) techniques to quantify correlations between composition, temperature, and material properties by analyzing high-throughput, in situ environmental photoluminescence (PL) experiments. Correlation heatmaps show the influence of Cs content on film degradation, and dimensionality reduction visualization methods uncover clear composition-based clusters despite overlapping datasets. A robust screening of 10 ML algorithms effectively forecasts PL features with single-composition, composition-generalized, and composition-generalized stacking approaches, with the highest-performing models achieving root mean squared errors of 1.84, 10.69, and 10.28, respectively. Using a multi-output composition-generalized Extra-Trees and Ridge Regression stacked model, a full PL spectrum can be predicted for any time, temperature, and composition input. Our ML-based framework could be expanded to other perovskite families, significantly reducing the analysis time to identify stable options for photovoltaics.