The conversion of carbon dioxide into clean fuels is regarded as an important route toward carbon neutrality. CO2 methanation, in particular, has drawn increasing interest due to its favorable thermodynamic properties and environmental benefits. Yet, large-scale deployment continues to face challenges such as insufficient catalyst activity at low temperatures and vulnerability to carbon deposition.

Researchers have now applied an explainable machine learning (ML) framework to support the rational design of nickel-based catalysts for CO2 methanation. Instead of relying on traditional trial-and-error methods, the study introduces a systematic approach for data processing, cross-validation, and ensemble learning model construction. Among the tested methods, a categorical boosting (CatBoost) model achieved R² values of 0.77 for CO2 conversion and 0.75 for CH4 selectivity.

By analyzing key descriptors, the study identified optimal reaction conditions: temperature between 250-350 °C, gas hourly space velocity below 15,000 cm³ g⁻¹ h⁻¹, BET surface area of 50-200 m² g⁻¹, and nickel content higher than 5%. These insights demonstrate how data-driven methods can guide catalyst optimization and shorten the pathway from laboratory research to industrial application.

"This work shows how machine learning can help us better understand the critical factors influencing CO2 methanation performance," said Hao Li, a Distinguished Professor at Tohoku University's Advanced Institute for Materials Research (WPI-AIMR. "By making the models explainable, we are not only predicting results but also gaining knowledge about why certain conditions matter."

Looking ahead, the research team will integrate density functional theory calculations and high-throughput experimental data to build multi-scale predictive models. They will also conduct systematic experimental validation to refine catalyst designs.

"Our goal is to establish a platform that combines computational chemistry, machine learning, and catalytic engineering," Li explained. "In doing so, we hope to contribute practical solutions for carbon recycling and the efficient use of renewable energy." This study provides a perspective on how explainable machine learning can be applied to catalyst research, supporting both the development of cleaner fuels and the broader transition to sustainable energy systems.

The study was published in the journal ACS Sustainable Chemistry & Engineering on August 22, 2025.