Integrating Machine Learning Insights in Membrane Electrode Assembly for CO ₂ Electrolysis

Abstract Membrane electrode assembly (MEA) electrolyzers hold promise for industrial‐scale CO 2 reduction applications. Almost all related studies are based on experiments; however, the multidimensionality and complexity of data present significant challenges to its optimization design. To promote the development of MEA electrolyzers from the perspective of data science, this work constructs an MEA electrolyzer device data dataset, which is composed of 501 devices from 204 relevant literature. By analyzing the data using statistical methods, the potential reasons for the device performance differences are identified. Subsequently, three indicators, including the 1st product, total current density, and the 1st product Faradaic efficiency, are selected and predicted by various machine learning (ML) algorithms. Random forest, gradient boosting, and support vector machine models are found to be optimal for each indicator, respectively. Guided by insights from interpretable ML analysis, an effective MEA electrolyzer featuring an Ag/C‐based catalyst was developed, which achieved a CO Faradaic efficiency of 100% at a current density of 200 mA cm −2 and stably operated at a cell voltage of 2.7 V over 100 h. Besides providing rich information and underlying interactions of the complicated parameters in MEA design, this work aims to motivate MEA optimization and its data management.