Prediction of magnetic Janus materials based on machine learning and first-principles calculations

The dataset in this study was processed and visualized using the pandas, matplotlib, and matminer libraries within the Scikit-learn framework. This dataset primarily includes data from the processes of machine learning model training, high-throughput material generation and screening, and first-principles calculation validation. The dataset consists of two main files: data_set and data of supplementary materials. The data_set file contains the original data (janus.csv), feature-engineered original data (janus_featured.csv), element-substituted data (prd_janus.csv), feature-engineered element-substituted data (prd_janus_featured.csv), and the model screening results (stable_high_magnetization.csv). The data of supplementary materials file includes Figure S1 and Table S1. Figure S1 presents an analysis of feature importance for lattice constants a = b, lattice constant c, formation energy, and magnetic moment categories during model training. Table S1 provides the formation energy and magnetic moment obtained from direct static self-consistent calculations for 13 unoptimized Janus structures.(1) janus.csv is the original dataset obtained from the Materials Project database, containing information on the chemical composition (elements and stoichiometry), crystal space group, lattice constants, formation energy, and total magnetic moment of 1,179 two-dimensional hexagonal ABC-type Janus materials.(2) janus_featured.csv is the dataset obtained by applying feature engineering based on elemental composition information to the original dataset.(3) prd_janus.csv is a dataset of 82,018 ABC-type two-dimensional Janus materials, not yet experimentally synthesized, generated by random substitution of elements A, B, and C from the periodic table based on the two-dimensional hexagonal ABC-type Janus structures in the original dataset.(4) prd_janus_featured.csv is the feature-engineered dataset of the element-substituted materials.(5) stable_high_magnetization.csv is the dataset obtained by applying a trained machine learning model to the feature-engineered element-substituted data, containing 4,204 Janus structures with lattice information, thermal stability, and high magnetic moment.