Capacity-prediction models for organic anode-active materials of lithium-ion batteries: advances in predictors using small data

Organic energy storage has attracted a lot of interest in enhancing performance and reducing the consumption of resources. If performance predictors are prepared, the exploration of new compounds can be accelerated without consumption of time, energy, and effort. In the present work, a new straightforward capacity predictor is constructed for the exploration of organic anode-active materials. Sparse modeling for small data (SpM-S) combining machine learning (ML) and our chemical insights was used to construct linear regression models of specific capacity. In our previous work, two predictors (models G1 and G2) were prepared using small datasets. However, the descriptors and prediction accuracy of these models were not validated. In the present work, a new improved model (model G3) has been constructed with the addition of new data. These three models were studied in terms of data science: namely, prediction accuracy, validity of the descriptors, amount of training data used, and effect of ML algorithms. The straightforward, generalizable, and interpretable model G3 can be applied to explore new organic anode-active materials. Moreover, these data-scientific approaches to model construction and validation can be used to explore new energy-related materials even with small data. A capacity prediction model for organic anode active materials was constructed using sparse modeling for small data. The new model was validated in terms of the prediction accuracy, validity of the descriptors, and amount of the training data.