In this study, machine learning (ML) models were employed to predict power density in bio-based supercapacitors based on electrode physical properties, elemental composition and potential window (PW) using repeated randomized seed splits and hyperparameter. Random forest (RF), extremely randomized tree (ERT) and categorical boosting (CB) ML algorithms were applied to build a predictive model from experimental data gathered from previously published research articles on bio-based supercapacitors. The model performance was evaluated using the coefficient of determination (R 2 ), root-mean-square error, mean absolute percentage error and mean absolute error (MAE). Among the evaluated models, CB algorithm outperformed others by achieving R 2 mean and standard deviation (SD) value of 0.729 ± 0.098, followed by ERT (0.664 ± 0.083) and RF (0.648 ± 0.087). Additionally, CB yielded the lowest MAE value of 2397.101 ± 4399.12 W/kg, followed by ERT (2854.871 ± 519.520 W/kg) and RF (2965.086 ± 565.380 W/kg). Shapley additive explanations analysis from CB model revealed that carbon-to-oxygen ratio (C/O) exhibited the most influential parameter for predicting power density, compared with other input variables. The results, when compared with key existing studies, demonstrated enhanced predictive capability and highlighted the potential of data-driven approaches for the design, prediction and interpretation of bio-based materials toward the development of sustainable supercapacitors. Therefore, this study shows an ML framework to predict the power density of bio-based supercapacitors based on electrode physical properties, elemental composition and PW. • Machine learning models for predicting power density of bio-based supercapacitors. • RF, ERT and CB were trained on literature data. • CB achieved best performance (R²=0.851) over ERT (R²=0.806) and RF (R²=0.722). • Shapley additive explanations show carbon-to-oxygen ratio is a major determinant. • CB can be used effectively in the prediction of power density for supercapacitor.
Edokpayi et al. (Wed,) studied this question.