Machine Learning as a Tool for Specific Capacity Prediction of Prospective Potassium Battery Electrodes

Application of data science and machine learning (ML) techniques in the domain of materials science has been increasing by leaps and bounds recently. With the help of ML, through input features derived from available databases we can rapidly screen materials based on our desired output. Capacity is one of the important parameters for choosing suitable electrode materials for high energy storage metal ion battery. Exploration of suitable electrode materials for metal ion batteries other than Li ion batteries (LIBs) has been deficient, though there is a need to develop alternative battery technologies with higher energy storage characteristics and environmental safety. In this work, we have considered Li, Na and K-ion electrode materials and their available battery data from Materials Project database to predict specific capacity of prospective K-ion battery electrode materials. Suitable features have been considered and developed to train the various ML algorithms. Mean Absolute Percentage Error has been considered as the performance metrics for assessment of different ML algorithms and among them, kernel ridge regression has been adopted as the most useful to predict the capacity of unknown electrode materials for K-ion battery. Using the value of specific capacity, the number of intercalated K ions in the formula unit of the non-intercalated electrode material compounds have also been calculated. DFT calculations have also been performed to verify the results obtained through ML. Our result shows ML is an encouraging alternative to computationally demanding DFT process as it can screen electrode materials rapidly for battery.