Generalizability and limitations of machine learning for yield prediction of oxidative coupling of methane



Siritanaratkul, Bhavin
(2022) Generalizability and limitations of machine learning for yield prediction of oxidative coupling of methane. Digital Chemical Engineering, 2. p. 100013.

Access the full-text of this item by clicking on the Open Access link.

Abstract

Product yields of catalytic reaction networks are dependent on many factors, encompassing both catalyst properties and reaction conditions. The oxidative coupling of methane (OCM) is a complex heterogeneous-homogeneous process, and the yield of the desired C2 products is non-linear with respect to reaction conditions. Herein, using two published datasets of OCM catalytic experimental results, I show that various machine learning (ML) algorithms can predict C2 yields from reaction conditions with a mean absolute error (MAE) of 0.5 – 1.0 percentage points in the best case. However, complications arising from real-world applications should be anticipated, therefore I investigated the effects of training set size, added noise, and out-of-sample partitions on the performance of ML algorithms. These results provide insights into the generalizability of the algorithms as well as caveats into the applicability of ML to reaction yield prediction

Item Type: Article
Uncontrolled Keywords: 46 Information and Computing Sciences, 4611 Machine Learning, Machine Learning and Artificial Intelligence
Divisions: Faculty of Science and Engineering > School of Physical Sciences
Depositing User: Symplectic Admin
Date Deposited: 02 May 2023 10:54
Last Modified: 21 Jun 2024 13:37
DOI: 10.1016/j.dche.2022.100013
Open Access URL: https://doi.org/10.1016/j.dche.2022.100013
Related URLs:
URI: https://livrepository.liverpool.ac.uk/id/eprint/3170088