Lattice parameters prediction of orthorhombic oxyhalides using machine learning

Abstract

Lattice parameters of orthorhombic oxyhalides with molecular formula AOX are predicted using KRR, LR, and GBR machine learning (ML) models. Seventeen data of orthorhombic oxyhalides are extracted from the Materials Project Database, and several features such as atomic radius, ionic radius, band gap, density, electro-negativity, and atomic mass are taken into account. After refining the data, they are used for ML training and testing processes. The actual values of the respective compounds' lattice parameters are compared with those predicted by different models. Then, the accuracy of their predictions is checked by calculating MAE, MSE, and R². The GBR model is more efficient in predicting lattice parameters 'b' and 'c', whereas KRR is found to be more more efficient in predicting 'a'. Further, using the random forest regression model, the features importance plot is also observed to understand which features play an important role in predicting the lattice parameters.