A Data-Driven Machine Learning Approach for Predictive Modeling of Transition Metal Dichalcogenide/Carbon Composite Supercapacitor Electrodes

Abstract

Transition metal dichalcogenides (MS2) materials possess unique pseudocapacitive characteristics but typically suffer from poor electrical conductivity and volume expansion during cycling, limiting their practical applications. Combining MS2 with conductive carbon materials has proved a common and effective strategy, yielding MS2/carbon composites with significantly improved supercapacitor performance. However, their development often relies on empirical trial-and-error methods, limiting systematic progress. Machine learning (ML) is revolutionizing materials science by enabling the rapid screening and prediction of material properties. In this study, we present a comprehensive ML framework to predict the electrochemical performances of MS2/carbon composite supercapacitor electrodes. Four ML models were evaluated, with the transformer-based TabPFN model achieving the highest predictive accuracy (R2 = 0.988, RMSE = 32.15 F g-1). SHapley Additive exPlanations (SHAP) identified covalent radius, specific surface area, and current density as critical factors governing the specific capacitance (Cs). Density functional theory (DFT) calculations were performed to evaluate the adsorption energies of potassium ions on various MS2 slabs, and the agreement with the ML results confirms the reliability of the ML predictions. This work establishes a data-driven ML approach to guide the design of advanced pseudocapacitive materials, significantly accelerating their development.