Machine learning to predict plasma-based CO2 conversion in dielectric barrier discharges

Abstract

Plasma-based CO 2 conversion is an emerging defossilization technology to combat climate change while producing valuable chemical feedstock, yet its optimization is hampered by complex nonlinear behavior and resource-intensive experimentation. In this work, we collected a comprehensive database, comprising 343 data points with four key operational and geometric parameters, published in literature between 2010 and 2025. Leveraging this dataset, we developed a hybrid machine learning (ML) model integrating artificial neural networks (ANN), random forest (RF) and extreme gradient boost (XGB) algorithms, to predict CO 2 conversion and energy efficiency (EE) in dielectric barrier discharge (DBD) reactors. Under a rigorous 5-fold cross validation protocol, the optimal hybrid model achieves an average R 2 of 0.867, while a non-cross validated train-test split yields R 2 of 0.917, with corresponding reductions in mean squared error relative to the best individual XGB model. Detailed residual and error distribution analyses show that the ensemble model substantially reduces extreme prediction errors, particularly for EE, thereby improving reliability across the full operating space. SHapley Additive exPlanations (SHAP) analysis further demonstrates that flow rate and power are the top two descriptors, accounting for approximately 70% of the model's prediction. This work establishes a robust, interpretable, and purely datadriven framework capable of unbiased predictions, while quantifying the generalizability of ML models in heterogeneous data environments, offering a practical tool to accelerate plasma-based gas conversion optimization.