Data-driven design and experimental validation of high-precision Ni–Co bimetallic compound-based pseudocapacitor models

Abstract

Bimetallic nickel–cobalt compounds are promising electrode materials for high-performance supercapacitors owing to their favorable physicochemical and electrochemical properties. However, optimization of such materials typically relies on extensive trial-and-error experiments, which are time-consuming and costly. This presents a significant challenge in accelerating the development of efficient electrode materials. Machine Learning (ML) offers a pathway to overcome this barrier by rapidly identifying key compositional and structural features from existing data. In this work, two complementary sets of performance descriptors were constructed based on the bidirectional performance considerations of electrode materials. By integrating high-quality experimental data from the literature on Ni–Co bimetallic supercapacitor electrodes, a dual-database architecture was established: one database compiles specific capacitance values, while the other contains corresponding cycling stability grades. Using this dual-performance framework, the study developed predictive models for the corresponding performance: an artificial neural network (ANN) regression model for specific capacitance and a gradient boosting classifier (GBC) for stability grading. This hybrid ML framework enables simultaneous prediction of numerical performance metrics and categorical durability ratings. The ANN regression model demonstrated high accuracy, achieving an R² of 0.92 on the test set for specific capacitance predictions. Guided by data-driven insights from SHAP ablation experiments and partial correlation analysis, the study identified the Ni–Co ratio and specific surface area as the two most influential descriptors. Following the modification directions indicated by these key factors, a NiCo₂O₄ sample was synthesized, achieving an ultra-high specific surface area of 1666 m² g⁻¹ and an excellent specific capacitance of 1538 F g⁻¹. The ANN-predicted capacitance agreed with experimental measurements to within 0.3%. Similarly, the GBC classifier accurately predicted the material's cycling retention grade as “B” after 10 000 cycles, consistent with experimental observations. Finally, the study further evaluated the model's generalization capability by using 14 newly collected and previously unseen real experimental samples as an independent validation set, reaffirming the model's robustness and predictive reliability. In summary, this work establishes two concise and effective bidirectional performance descriptors that integrate scientifically grounded electrode material attributes with explicit electrochemical indicators to model and characterize the pseudocapacitance and cycling stability of Ni–Co bimetallic compounds. A representative dual-transition-metal electrode database and hybrid modeling framework were constructed based on these descriptors. Through data-driven variable analysis, the experimental design was effectively guided, and the performance characterization of the synthesized NiCo₂O₄ sample validated both the predictive accuracy of the ANN model and the reliability of the GBC model. These results demonstrate the models' capability to rapidly assess the feasibility of experimental schemes based on controllable descriptor combinations, thereby accelerating the development of advanced electrode materials. This iterative, data-driven design strategy—integrating ML-guided synthesis with experimental feedback—offers broad implications for the rational design of high-performance supercapacitor materials.