Predictive Modeling and SHAP (SHapley Additive ExPlanations) Analysis for Enhancing Natural Dye-Sensitized Solar Cell Performance

Abstract

Achieving high power conversion efficiency (PCE) in natural dye-sensitized solar cells remains a challenge. To better understand such challenges and explore potential solutions, a dataset is created from 113 experimental articles published recently. The data are analyzed using random forest and gradient boosting algorithms, and predictive models for open-circuit voltage (V_oc), short-circuit current density (J_sc), fill factor (FF), and PCE are developed. The model predictions are quite successful for all four performance indicators, with root mean square errors of 0.1, 1.7, 0.09, and 0.5 for V_oc, J_sc, FF, and PCE, respectively. The SHAP (SHapley Additive exPlanations) analysis is also performed to determine the effects of the descriptors on output variables. It is found that the dye extraction (such as dye/solvent ratio and extraction time) and deposition methods are highly influential for all four performance variables. It is also observed that chlorophyll, anthocyanin, and carotenoid dyes can improve V_oc, whereas there is no major dye type that can be identified for improvement of J_sc. Flavonoids, curcumin, and tannins dyes are found to be capable of increasing the cell FF; only the anthocyanin and chlorophyll can have a direct positive impact on the PCE output.