Machine-Learning-Assisted\nDescriptors Identification\nfor Indoor Formaldehyde Oxidation Catalysts

Abstract

The\ndevelopment of highly efficient catalysts for formaldehyde\n(HCHO) oxidation is of significant interest for the improvement of\nindoor air quality. Up to 400 works relating to the catalytic oxidation\nof HCHO have been published to date; however, their analysis for collective\ninference through conventional literature search is still a challenging\ntask. A machine learning (ML) framework was presented to predict catalyst\nperformance from experimental descriptors based on an HCHO oxidation\ncatalysts database. MnO_<i>x</i>, CeO₂, Co₃O₄, TiO₂, FeO_<i>x</i>, ZrO₂, Al₂O₃, SiO₂, and carbon-based catalysts with different promoters were compiled\nfrom the literature. Notably, 20 descriptors including reaction catalyst\ncomposition, reaction conditions, and catalyst physical properties\nwere collected for data mining (2263 data points). Furthermore, the\neXtreme Gradient Boosting algorithm was employed, which successfully\npredicted the conversion efficiency of HCHO with an R-square value\nof 0.81. Shapley additive analysis suggested Pt/MnO₂ and\nAg/Ce–Co₃O₄ exhibited excellent catalytic\nperformance of HCHO oxidation based on the analysis of the entire\ndatabase. Validated by experimental tests and theoretical simulations,\nthe key descriptor identified by ML, i.e., the first promoter, was\nfurther described as metal–support interactions. This study\nhighlights ML as a useful tool for database establishment and the\ncatalyst rational design strategy based on the importance of analysis\nbetween experimental descriptors and the performance of complex catalytic\nsystems.