eprintid: 19699
rev_number: 2
eprint_status: archive
userid: 1
dir: disk0/00/01/96/99
datestamp: 2024-06-04 14:19:26
lastmod: 2024-06-04 14:19:26
status_changed: 2024-06-04 14:15:38
type: article
metadata_visibility: show
creators_name: Abdullah, N.H.
creators_name: Zaini, D.
creators_name: Lal, B.
title: Prediction of ionic liquids toxicity using machine learning models for application to gas hydrate
ispublished: pub
keywords: Gas hydrates; Gas industry; Hydration; Ionic liquids; Learning systems; Machine learning; Mean square error, Descriptors; Extra-trees; Gradient boost; LASSO; Machine learning models; Oil and Gas Industry; Predictive performance; Random forests; Ridge; Support vector regressions, Decision trees
note: cited By 0
abstract: Ionic liquids (ILs) are highly favored in the oil and gas industry as gas hydrate inhibitors due to their dual functionality as thermodynamic inhibitor and kinetic hydrate inhibitor. Though known as the �green alternatives,� concerns about the effects of ILs in the environment are rising such that ILs can stabilize in water systems. Furthermore, there are insufficient data on the toxicity of ILs, limiting the use of ILs for industrial applications. Ridge, LASSO, decision tree, random forest, extra tree, gradient boost, and support vector regressions were used to develop IL toxicity predictive models. Random forest yielded the strongest predictive performance, scoring the highest R2 value of 0.86, with mean absolute error and root mean square error values of 0.32 and 0.43, respectively. Feature selections were conducted to investigate the contributions of the five molecular descriptors involved in developing regression models in this work. Descriptor MSDC was found to contribute the highest at 67 in predicting the toxicity of ILs, followed by SNarA and MAXDPC, demonstrating contributions of 15.2 and 14.1, respectively. Further quantitative structure�activity relationship model validations were executed; the use of three descriptors resulted in a 2 increase in predictive performance for decision tree regression, whereas R2 values remained the same for random forest, extra tree, and gradient boosting. © 2024 American Institute of Chemical Engineers.
date: 2024
publisher: John Wiley and Sons Inc
official_url: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85189079520&doi=10.1002%2fprs.12599&partnerID=40&md5=c30a79a02cfb7b1d15b52063a7d804bd
id_number: 10.1002/prs.12599
full_text_status: none
publication: Process Safety Progress
volume: 43
number: S1
pagerange: S199-S212
refereed: TRUE
issn: 10668527
citation:   Abdullah, N.H. and Zaini, D. and Lal, B.  (2024) Prediction of ionic liquids toxicity using machine learning models for application to gas hydrate.  Process Safety Progress, 43 (S1).  S199-S212.  ISSN 10668527