Leveraging Interpretable Features and Clustering Agreements for Assistive Materials Discovery

Data science is intrinsically interdisciplinary; however, end-users of machine learning models are not always trained data scientists. For this reason, it is crucial that these models be infused with domain knowledge to increase explainability and trust in their output. Our goal is to assign domain-aware confidence scores to models' outputs to help experts make informed decisions. Our hypothesis is that given consensus-based confidence scores, end-users will be more willing to trust and thus adopt machine learning models. We test this hypothesis in materials informatics, a field that has the potential to greatly reduce time-to-market and development costs for new materials as it leverages machine learning and large datasets for targeted materials design. Automated phase-mapping seeks to discover the relationship between materials synthesis and materials phases, as materials with similar structure are likely to have similar properties and vice versa. A particular challenge is identifying composition-phase regions, i.e., regions of composition space where materials are composed of similar lattice-structure phases. This is challenging because structure measurements per sample are high dimensional and suffer from the curse of dimensionality, making results difficult to interpret and generalize. Towards our goal, we quantify agreements (or disagreements) across carefully chosen clustering methods used to discover composition-phase regions by defining confidence (or uncertainty) scores to identify the correct number of phase regions present in X-Ray Diffraction (XRD) datasets. Building a series of clustering experiments, we show that we can narrow in on the correct number of phase regions in two ternary composition spreads.