2021/07/08 by Rasmus Tranås, Ole Martin Løvvik, Tranås, Rasmus +5 · 2 citations
Materials Science · #Advanced Thermoelectric Materials and Devices #FOS: Physical sciences #Machine Learning in Materials Science #Materials Science (cond-mat.mtrl-sci) #Thermal properties of materials
paper · pdf · doi:10.48550/arxiv.2107.03735
openalex publication_date 2021/07/08 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Low lattice thermal conductivity is essential for high thermoelectric\nperformance of a material. Lattice thermal conductivity is often computed using\ndensity functional theory (DFT), typically at a high computational cost.\nTraining machine learning models to predict lattice thermal conductivity could\noffer an effective procedure to identify low lattice thermal conductivity\ncompounds. However, in doing so, we must face the fact that such compounds can\nbe quite rare and distinct from those in a typical training set. This\ndistinctness can be problematic as standard machine learning methods are\ninaccurate when predicting properties of compounds with features differing\nsignificantly from those in the training set. By computing the lattice thermal\nconductivity of 122 half-Heusler compounds, using the temperature-dependent\neffective potential method, we generate a data set to explore this issue. We\nfirst show how random forest regression can fail to identify low lattice\nthermal conductivity compounds with random selection of training data. Next, we\nshow how active selection of training data using feature and principal\ncomponent analysis can be used to improve model performance and the ability to\nidentify low lattice thermal conductivity compounds. Lastly, we find that\nactive learning without the use of DFT-based features can be viable as a\nquicker way of selecting samples.\n