Machine Learning-Based Prediction of Supercapacitor Capacitance for MgCo₂O₄ Electrodes.

Area of Science:

• Materials Science
• Electrochemistry
• Machine Learning

Background:

• Electrode materials critically influence supercapacitor cost and performance.
• Accurate capacitance prediction is vital for advancing supercapacitor technology.
• Magnesium cobalt oxide (MgCo2O4) exhibits high theoretical capacitance, making it a promising electrode material.

Purpose of the Study:

• To develop a highly accurate machine learning model for predicting the capacitance of MgCo2O4 electrode materials.
• To identify optimal input features for capacitance prediction using feature selection techniques.
• To establish a reliable computational tool for guiding experimental design in supercapacitor research.

Main Methods:

• Extensive data extraction from published literature on MgCo2O4.
• Application of Recursive Feature Elimination (RFE) with Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Regression Tree (RT) for feature selection.
• Development and evaluation of nine machine learning models by combining feature subsets with regression algorithms.

Main Results:

• The XGBoost-Recursive Feature Elimination-XGBoost (XGB-RFE-XGB) model demonstrated superior performance.
• The XGB-RFE-XGB model achieved a coefficient of determination (R-squared) of 0.95.
• Excellent predictive accuracy was confirmed by low Root Mean Squared Error (RMSE) of 111.83 F/g and Mean Absolute Error (MAE) of 68.25 F/g.

Conclusions:

• The XGB-RFE-XGB model is a stable and reliable tool for predicting MgCo2O4 capacitance.
• This predictive capability can significantly accelerate the development and optimization of supercapacitors.
• The study highlights the potential of machine learning in materials discovery and electrochemical device engineering.