Machine learning-guided optimization, predictive modeling, and experimental validation of MXene-based supercapacitors

초록

In the present work, we extracted approximately 7,500 data points related to MXene-based supercapacitors (SCs) from the existing literature and categorized them based on galvanostatic charge-discharge (GCD) and cyclic voltammetry (CV) measurements. The datasets are analyzed using classification and regression trees to gain electrochemical insights. The feature importance analysis revealed that the working electrode's material composition, substrate, electrolyte, counter electrode, etching time, etchant, fabrication technique, potential window, and current density are the most influential parameters governing the SC performance. Moreover, different ML algorithms such as random forest (RF), artificial neural network (ANN), gradient boosting (GB), and linear model (LM) are utilized to predict the specific capacitance, cyclic stability, and capacitive retention of MXene-based SCs. Among them, RF showed the highest predictive accuracy (Pearson's r ti 0.94, Adj. R2 ti 0.88), outperforming GB (r ti 0.56, R2 ti 0.31), ANN (r ti 0.44, R2 ti 0.17), and LM (r ti 0.29, R2 ti 0.06) across both CV and GCD datasets. The optimized ML model achieved prominent prediction accuracy, suggesting strong agreement with reported experimental data. This study offers a reliable and efficient approach to optimizing MXenebased SCs and accelerates the design of advanced materials for energy storage applications.

키워드