Ensemble Approach Assisted Specific Capacitance Prediction for Heteroatom-Doped High-Performance Supercapacitors

Abstract

Specific capacitance plays a critical role when assessing the performance of a supercapacitor. Hence, its prediction is crucial for evaluating the electrochemical performance of electric double-layer capacitors (EDLCs). Machine learning (ML) offers the prospect of predicting capacitance with nominal investment in the synthesis and testing of electrode materials. Herein, six ML models: random forest (RF), artificial neural network (ANN), random tree (RT), random committee (RC), random subspace (RS), and support vector machine (SVM) regressor are used to analyze the effect of four hetero atom doping (nitrogen, boron, sulfur, and phosphorous) on the electrochemical performance of EDLCs. Amongst all, RF, ANN, and RS showed the highest correlation values of 0.9996, 0.9993 and 0.9867, respectively, and the lowest root mean square values of 0.93, 1.19, and 2.31, respectively, through selection of 12 key input descriptors on the basis of physical, structural, test, operational, and doping parameters. Furthermore, attribute prioritization was introduced to identify and rank important features within the dataset. It highlights that specific surface area, total pore volume, and nitrogen are the most significant descriptors among 12 selected input features. With fewer iterations, the developed models’ estimation accuracy surpassed other state-of-art models in literature. In perspective, this study considers an extensive dataset extracted from more than 250 research articles on heteroatom-doped carbon electrodes. It also provides insights into the significance of ML modeling on the electrochemical technology.