Machine Learning Enabled Screening of Single Atom Alloys: Predicting Reactivity Trend for Ethanol Dehydrogenation

A machine learning (ML) approach implementing the gradient boosting regressor (GBR) algorithm is applied to predict the binding energies of oxygen (EO) and carbon (EC) atoms on single atom alloys (SAAs) of Cu, Ag and Au. Readily available periodic properties of the transition metals are utilized as input features in the model. Their relative contribution in adsorbate‐metal interaction is assessed to develop a comprehensive descriptor. In test runs, the ML model is observed to predict EO and EC with significantly reduced errors (∼0.2 eV). Further, ML approach is augmented with an ab initio microkinetic model (MKM) for non‐oxidative dehydrogenation (NODH) of ethanol. The ML‐MKM is calculated to yield higher turnover frequency for ethanol conversion on NiAu, NiAg, and PtAg SAAs as compared to their monometallic counterparts Au and Ag catalysts. Overall, the ML model provides a rationale for feature selection to synthesize catalytically active SAAs of group 11 elements using fast‐track in silico screening. Single‐Atoms for Catalysis: A machine learning approach is implemented to predict the oxygen (EO) and carbon (EC) binding energies on single atom alloys (SAAs) for fast‐track screening in ethanol dehydrogenation reaction.