Novelty detection in the design of synthesis of garnet-structured solid electrolytes

Abstract

Recent decades have shown arising growth-on-demand fusion of machine learning into all of the areas of chemistry and materials science. In this study, we examine one aspect of using these technologies to gain an advantage in the search for new knowledge in increased amounts of experimental data. The novelty detection approaches are aimed to identify the data artefacts. The analysis of such artefacts, “outliers”, in details of synthesis of garnet-structured solid electrolytes was chosen as an example of the practical application of this methodology. Particular attention was paid to the choice of the precursors. The thermodynamic data such as the heat of formation from the pure oxides as well as the results of drop solution calorimetry for simple oxides were involved as the descriptors for the studied systems. The overall performance of novelty/outlier detection of “outliers” was characterized using the AUC (area under the curve) value and assessed to be 0.71 – 0.72 varying the complexity of data description. It was found that all “outlier” due to the precursors choice compounds were successfully identified. The regression analysis was performed to analyze the impact of outlier/novelty detection on model predictive performance as well as to elucidate the relationship with the data diversity and the complexity of data description.