Application of machine learning methods for the prediction of crystal system of cathode materials in lithium-ion batteries

The system of crystal structure has a major effect on the physical and chemical properties of Li-ion silicate cathodes. Hence, the prediction of crystal system has a vital importance to estimate many other properties of cathodes for applications in batteries. Three major crystal systems (monoclinic, orthorhombic and triclinic) of silicate-based cathodes with Li–Si–(Mn, Fe, Co)–O compositions were predicted using wide range of classification algorithms in machine learning. The calculations are based on the results of density functional theory calculations from Materials Project. The strong correlation between the crystal system and other physical properties of the cathodes was confirmed based on the feature evaluation in the statistical models. In addition, the parameters of various classification methods were optimized to obtain the best accuracy of prediction. Ensemble methods including random forests and extremely randomized trees provided the highest accuracy of prediction among other classification methods in the Monte Carlo cross validation tests.