Prediction of lattice parameters of tetragonal oxyhalides AOX

Abstract

Machine learning enables computers to emulate human intelligence for complex data analysis and pattern
recognition. This work utilizes machine learning to predict lattice parameters in tetragonal oxyhalide compounds with molecular formula AOX. Four supervised learning methods - random forest regression, gradient boosting regression, support vector regression, and kernel ridge regression - are employed to forecast lattice parameters from features including atomic radii, ionic radii, atomic masses, electronegativities, band gaps, formation energies, and densities. Model accuracy is evaluated using mean absolute error and R² as regression scoring measures. An analysis of gradient boosting regression determines the predictive capacity of distinct features toward lattice parameters. Comparisons identify kernel ridge regression as optimal for predicting lattice constant a, with the highest R² of 0.840; whereas gradient boosting shows superior in modeling lattice parameter c with a maximum R² reaching 0.948. This research demonstrates the successful application of machine learning methodologies for predicting material properties, enabling the estimation of lattice parameters in tetragonal oxyhalides.