|About this Abstract
|2022 TMS Annual Meeting & Exhibition
|AI/Data Informatics: Computational Model Development, Validation, and Uncertainty Quantification
|Applicability Domain for Prediction Models of Thermoelectric Properties Based on Similarity to Known Materials
|Masaya Kumagai, Yukari Katsura, Yuki Ando, Atsumi Tanaka, Koji Tsuda, Ken Kurosaki
|On-Site Speaker (Planned)
Machine learning (ML) has attracted attention for material exploration and design. In general, randomly extracted subsets as unknown materials are used to evaluate the performance of ML models. However, in many cases, the details of subsets are not mentioned. Therefore, it is unclear how extent the model is applicable to unknown materials in the actual vast material space. In this study, we constructed ML models to predict thermoelectric properties from feature vectors based on chemical composition, and evaluated the performance of the ML models after clarifying the unknown materials by the following two ways: 1. material family classification of unknown materials based on clustering of known materials, 2. determination of applicability domain for the ML models based on its similarity to known materials.
|Machine Learning, Energy Conversion and Storage,