Data-Driven Design and Understanding of Noble Metal-Based Water–Gas Shift Catalysts from Literature Data

Catalyst informatics and catalyst design have the potential
to
facilitate and speed up catalyst discovery, as this is a complex undertaking
involving variables associated with the catalysts themselves and operating
conditions. Herein, a Machine Learning (ML)-assisted methodology coupled
with data visualization to design descriptors for catalyst materials
are proposed using a previously reported literature data set of the
Water–Gas Shift (WGS) reaction. This entails two different
approaches to represent catalysts as part of the input and propose
catalysts based on their predicted CO conversion. The analysis covers
the design of the descriptors employed by the models, as well as the
results of an inverse prediction, that uncovered potential catalysts
that can be researched for high CO conversion (≥95%), with
Random Forest Regression predicting promoted Au/CeO2–ZrO2 and Support-Vector Regression predicting promoted Au/CeO2, Ru/CeO2, and Rh/CeO2 as the best overall
catalyst candidates, and Yb/Au/CeO2–ZrO2 to be of interest for WGS applications at low temperatures.