Ramanathan Eshwar S, Chowdhury Chandra
Department of Ocean Engineering, Indian Institute of Technology Madras, Chennai, 600036, India.
Institute of Catalysis Research and Technology (IKFT), Karlsruhe Institute of Technology (KIT), 76344, Eggeinstein-Leopoldshafen, Germany.
Chemphyschem. 2023 Nov 2;24(21):e202300308. doi: 10.1002/cphc.202300308. Epub 2023 Sep 25.
The growing number of studies and interest in two-dimensional (2D) materials has not yet resulted in a wide range of material applications. This is a result of difficulties in getting the properties, which are often determined through numerical experiments or through first-principles predictions, both of which require lots of time and resources. Here we provide a general machine learning (ML) model that works incredibly well as a predictor for a variety of electronic and structural properties such as band gap, fermi level, work function, total energy and area of unit cell for a wide range of 2D materials derived from the Computational 2D Materials Database (C2DB). Our predicted model for classification of samples works extraordinarily well and gives an accuracy of around 99 %. We are able to successfully decrease the number of studied features by employing a strict permutation-based feature selection method along with the sure independence screening and sparsifying operator (SISSO), which further supports the design recommendations for the identification of novel 2D materials with the desired properties.
对二维（2D）材料的研究及兴趣日益增长，但尚未带来广泛的材料应用。这是由于获取材料特性存在困难，这些特性通常通过数值实验或第一性原理预测来确定，而这两者都需要大量时间和资源。在此，我们提供了一个通用的机器学习（ML）模型，它作为多种电子和结构特性的预测器表现出色，这些特性包括带隙、费米能级、功函数、总能量以及源自计算二维材料数据库（C2DB）的广泛二维材料的晶胞面积。我们的样本分类预测模型效果极佳，准确率约为99%。通过采用基于严格排列的特征选择方法以及确定性独立筛选和稀疏化算子（SISSO），我们成功减少了研究特征的数量，这进一步支持了针对具有所需特性的新型二维材料识别的设计建议。
