Unlocking the potential of Ti3C2 electrodes: a data-driven capacitance prediction study

Abstract

In the dynamic realm of energy storage devices, supercapacitors (SCs) stand out as promising options. Among the various materials considered for SC electrodes, layered substances like Ti₃C₂ have drawn attention due to their outstanding electrochemical qualities, especially their impressive volumetric capacitance. This study focuses on assessing the predictive abilities of three machine learning models: Bayesian ridge regression (BRR), K-nearest neighbors (KNN), and artificial neural network (ANN) in estimating specific capacitance in Ti₃C₂-based supercapacitors. BRR offered reliable predictions with an R-squared (R²) value of 0.759 and a low root mean square error (RMSE) of 0.074. KNN excelled in predicting supercapacitor performance with an impressive R² of 0.928 and a minimal RMSE of 0.040. However, the ANN model stood out as it could reveal the significance of various inputs much like the human brain's intricate functioning. It achieved a high R² of 0.8929 with a low RMSE of 0.0493, demonstrating its proficiency in capturing complex relationships in the dataset. The precise tuning of hyperparameters further enhanced its accuracy. The use of SHAP (SHapley Additive exPlanations) values emphasized cation mobility, and scan rates as key contributing factors. These findings provide a strong foundation for utilizing machine learning to predict specific capacitance in Ti₃C₂-based supercapacitors. Researchers can benefit from these versatile tools for precise predictions, facilitating systematic supercapacitor design and enhancing our understanding of electrode materials.