A comparative analysis of machine learning predictive models for the oxidative coupling of methane reaction

Lord Ugwu; Yasser Morgan; Hussameldin Ibrahim

doi:10.1016/j.mtcomm.2024.111446

A comparative analysis of machine learning predictive models for the oxidative coupling of methane reaction

Lord Ugwu, Yasser Morgan, Hussameldin Ibrahim

Department of Systems and Enterprises

Stevens Institute of Technology

Research output: Contribution to journal › Article › peer-review

Abstract

The amalgamation of catalytic and electronic characteristics, along with empirical data, furnishes enhanced insights into catalyst analysis, thereby advancing innovation and the design of heterogeneous catalytic reactions. In this research, we juxtaposed the catalysts' electronic properties, including the Fermi energy, bandgap energy, and magnetic moment of catalyst components with available high-throughput OCM experimental data, to prognosticate catalytic efficacy and the resultant reaction outcomes, encompassing methane conversion and yields of ethylene, ethane, and carbon dioxide yields. Comparative evaluation of diverse machine learning models indicates that the extreme gradient boost regression model stands out for its superior predictive accuracy in evaluating catalytic performance with an average R² of 0.91. The order of performance of the modeling techniques was XGBR > RFR > DNN > SVR. The MSE and MAE of the XGBR models appeared to be lower than those of the other modeling techniques, with numbers ranging from 0.26 to 0.08 for MSE and 1.65–0.17 for MAE. The MSE and MAE of the models trained with a particular dataset generally aligned with those of an external dataset not seen by the model at the time of training or testing, as well as the MSE from bootstrap sampling, confirming the high generalizability of the models. The accuracy of the models was in the order of C₂H₆y > CH₄_conv > CO₂y > C₂y > C₂H₄y. Comparing the quantification of uncertainty of the RCCEP feature-based predictive models of the different ML techniques based on the prediction band suggests that the ML techniques rank in the order of XGBR > RFR > DNN > SVR. In analyzing the impact of the model features, the combined ethylene and ethane yield increases with an increase in dataset features, including the number of moles of the alkali/alkali-earth metal in the catalyst, the atomic number of the catalyst promoter and the Fermi energy of the metal, and just relatively in the case of temperature, suggesting a highly non-linear relationship between the combined ethylene and ethane yield and temperature. The extent of the impact of these features on the predictive model for the combined ethylene and ethane yield was 5.91 %, 13.28 % and 33.76 % for the atomic number of the promoter, the number of moles of the alkali/alkali-earth metal in the catalyst and the reaction temperature, respectively. Other features, including the bandgap of the active metal oxide and the support, as well as the Fermi energy of the catalyst support, were also seen to have a relatively modest impact on the predictive models for the combined ethylene and ethane yield and methane conversion.

Original language	English
Article number	111446
Journal	Materials Today Communications
Volume	42
DOIs	https://doi.org/10.1016/j.mtcomm.2024.111446
State	Published - Jan 2025

ASJC Scopus subject areas

General Materials Science
Mechanics of Materials
Materials Chemistry

Keywords

Ethylene
Extreme gradient boost regression
Machine learning
Oxidative coupling of methane

Access to Document

10.1016/j.mtcomm.2024.111446

Cite this

@article{694979274a12424e8c534d50018e2721,

title = "A comparative analysis of machine learning predictive models for the oxidative coupling of methane reaction",

abstract = "The amalgamation of catalytic and electronic characteristics, along with empirical data, furnishes enhanced insights into catalyst analysis, thereby advancing innovation and the design of heterogeneous catalytic reactions. In this research, we juxtaposed the catalysts' electronic properties, including the Fermi energy, bandgap energy, and magnetic moment of catalyst components with available high-throughput OCM experimental data, to prognosticate catalytic efficacy and the resultant reaction outcomes, encompassing methane conversion and yields of ethylene, ethane, and carbon dioxide yields. Comparative evaluation of diverse machine learning models indicates that the extreme gradient boost regression model stands out for its superior predictive accuracy in evaluating catalytic performance with an average R2 of 0.91. The order of performance of the modeling techniques was XGBR > RFR > DNN > SVR. The MSE and MAE of the XGBR models appeared to be lower than those of the other modeling techniques, with numbers ranging from 0.26 to 0.08 for MSE and 1.65–0.17 for MAE. The MSE and MAE of the models trained with a particular dataset generally aligned with those of an external dataset not seen by the model at the time of training or testing, as well as the MSE from bootstrap sampling, confirming the high generalizability of the models. The accuracy of the models was in the order of C2H6y > CH4_conv > CO2y > C2y > C2H4y. Comparing the quantification of uncertainty of the RCCEP feature-based predictive models of the different ML techniques based on the prediction band suggests that the ML techniques rank in the order of XGBR > RFR > DNN > SVR. In analyzing the impact of the model features, the combined ethylene and ethane yield increases with an increase in dataset features, including the number of moles of the alkali/alkali-earth metal in the catalyst, the atomic number of the catalyst promoter and the Fermi energy of the metal, and just relatively in the case of temperature, suggesting a highly non-linear relationship between the combined ethylene and ethane yield and temperature. The extent of the impact of these features on the predictive model for the combined ethylene and ethane yield was 5.91 %, 13.28 % and 33.76 % for the atomic number of the promoter, the number of moles of the alkali/alkali-earth metal in the catalyst and the reaction temperature, respectively. Other features, including the bandgap of the active metal oxide and the support, as well as the Fermi energy of the catalyst support, were also seen to have a relatively modest impact on the predictive models for the combined ethylene and ethane yield and methane conversion.",

keywords = "Ethylene, Extreme gradient boost regression, Machine learning, Oxidative coupling of methane",

author = "Lord Ugwu and Yasser Morgan and Hussameldin Ibrahim",

note = "Publisher Copyright: {\textcopyright} 2025 The Authors",

year = "2025",

month = jan,

doi = "10.1016/j.mtcomm.2024.111446",

language = "English",

volume = "42",

journal = "Materials Today Communications",

issn = "2352-4928",

publisher = "Elsevier Ltd",

}

TY - JOUR

T1 - A comparative analysis of machine learning predictive models for the oxidative coupling of methane reaction

AU - Ugwu, Lord

AU - Morgan, Yasser

AU - Ibrahim, Hussameldin

PY - 2025/1

Y1 - 2025/1

N2 - The amalgamation of catalytic and electronic characteristics, along with empirical data, furnishes enhanced insights into catalyst analysis, thereby advancing innovation and the design of heterogeneous catalytic reactions. In this research, we juxtaposed the catalysts' electronic properties, including the Fermi energy, bandgap energy, and magnetic moment of catalyst components with available high-throughput OCM experimental data, to prognosticate catalytic efficacy and the resultant reaction outcomes, encompassing methane conversion and yields of ethylene, ethane, and carbon dioxide yields. Comparative evaluation of diverse machine learning models indicates that the extreme gradient boost regression model stands out for its superior predictive accuracy in evaluating catalytic performance with an average R2 of 0.91. The order of performance of the modeling techniques was XGBR > RFR > DNN > SVR. The MSE and MAE of the XGBR models appeared to be lower than those of the other modeling techniques, with numbers ranging from 0.26 to 0.08 for MSE and 1.65–0.17 for MAE. The MSE and MAE of the models trained with a particular dataset generally aligned with those of an external dataset not seen by the model at the time of training or testing, as well as the MSE from bootstrap sampling, confirming the high generalizability of the models. The accuracy of the models was in the order of C2H6y > CH4_conv > CO2y > C2y > C2H4y. Comparing the quantification of uncertainty of the RCCEP feature-based predictive models of the different ML techniques based on the prediction band suggests that the ML techniques rank in the order of XGBR > RFR > DNN > SVR. In analyzing the impact of the model features, the combined ethylene and ethane yield increases with an increase in dataset features, including the number of moles of the alkali/alkali-earth metal in the catalyst, the atomic number of the catalyst promoter and the Fermi energy of the metal, and just relatively in the case of temperature, suggesting a highly non-linear relationship between the combined ethylene and ethane yield and temperature. The extent of the impact of these features on the predictive model for the combined ethylene and ethane yield was 5.91 %, 13.28 % and 33.76 % for the atomic number of the promoter, the number of moles of the alkali/alkali-earth metal in the catalyst and the reaction temperature, respectively. Other features, including the bandgap of the active metal oxide and the support, as well as the Fermi energy of the catalyst support, were also seen to have a relatively modest impact on the predictive models for the combined ethylene and ethane yield and methane conversion.

AB - The amalgamation of catalytic and electronic characteristics, along with empirical data, furnishes enhanced insights into catalyst analysis, thereby advancing innovation and the design of heterogeneous catalytic reactions. In this research, we juxtaposed the catalysts' electronic properties, including the Fermi energy, bandgap energy, and magnetic moment of catalyst components with available high-throughput OCM experimental data, to prognosticate catalytic efficacy and the resultant reaction outcomes, encompassing methane conversion and yields of ethylene, ethane, and carbon dioxide yields. Comparative evaluation of diverse machine learning models indicates that the extreme gradient boost regression model stands out for its superior predictive accuracy in evaluating catalytic performance with an average R2 of 0.91. The order of performance of the modeling techniques was XGBR > RFR > DNN > SVR. The MSE and MAE of the XGBR models appeared to be lower than those of the other modeling techniques, with numbers ranging from 0.26 to 0.08 for MSE and 1.65–0.17 for MAE. The MSE and MAE of the models trained with a particular dataset generally aligned with those of an external dataset not seen by the model at the time of training or testing, as well as the MSE from bootstrap sampling, confirming the high generalizability of the models. The accuracy of the models was in the order of C2H6y > CH4_conv > CO2y > C2y > C2H4y. Comparing the quantification of uncertainty of the RCCEP feature-based predictive models of the different ML techniques based on the prediction band suggests that the ML techniques rank in the order of XGBR > RFR > DNN > SVR. In analyzing the impact of the model features, the combined ethylene and ethane yield increases with an increase in dataset features, including the number of moles of the alkali/alkali-earth metal in the catalyst, the atomic number of the catalyst promoter and the Fermi energy of the metal, and just relatively in the case of temperature, suggesting a highly non-linear relationship between the combined ethylene and ethane yield and temperature. The extent of the impact of these features on the predictive model for the combined ethylene and ethane yield was 5.91 %, 13.28 % and 33.76 % for the atomic number of the promoter, the number of moles of the alkali/alkali-earth metal in the catalyst and the reaction temperature, respectively. Other features, including the bandgap of the active metal oxide and the support, as well as the Fermi energy of the catalyst support, were also seen to have a relatively modest impact on the predictive models for the combined ethylene and ethane yield and methane conversion.

KW - Ethylene

KW - Extreme gradient boost regression

KW - Machine learning

KW - Oxidative coupling of methane

UR - http://www.scopus.com/inward/record.url?scp=85213882164&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85213882164&partnerID=8YFLogxK

U2 - 10.1016/j.mtcomm.2024.111446

DO - 10.1016/j.mtcomm.2024.111446

M3 - Article

SN - 2352-4928

VL - 42

JO - Materials Today Communications

JF - Materials Today Communications

M1 - 111446

ER -

A comparative analysis of machine learning predictive models for the oxidative coupling of methane reaction

Abstract

ASJC Scopus subject areas

Keywords

Access to Document

Other files and links

Fingerprint

Cite this