Accelerating catalytic experimentation of water gas shift reaction using machine learning models

Abstract

Catalyst is an essential component of any reaction pathway, and in former times, “trial and error” approaches were used to identify viable candidates for a chosen reaction. Our study is built on the use of machine learning (ML) model, to reduce the time and cost involved in the catalyst screening process. Here we focussed on a probe reaction i.e. Water Gas Shift Reaction (WGSR) to test the developed model and subsequently predict suitable catalysts. Firstly, available experimental data from the literature were collected and represented using the Sorted Weighted Elemental Descriptor (SWED) and conventional techniques. A standard ML algorithm was used to develop the model, and after a 10-fold validation and RMSE evaluation, the best model was used to predict the new potential candidates for the chosen reaction. The pattern has been explored by embarking on feature importance studies, and finally, to augment with practicality, a user interface is created.