ISSN print edition: 0366-6352 ISSN electronic edition: 1336-9075 Registr. No.: MK SR 9/7 Published monthly

Machine learning frameworks to accurately compute CO₂ capture in porous liquids

Farag M. A. Altalbawy, Soumya V. Menon, Anupam Yadav, Subhashree Ray, Kapil Ghai, Atreyi Pramanik, Shirin Shomurotova, Gunjan Garg, Ahmad Alkhayyat, and Ahmad Abumalek

Department of Chemistry, University College of Duba, University of Tabuk, Tabuk, Saudi Arabia

E-mail: ahmedalkhayyat85@gmail.com

Received: 9 October 2025  Accepted: 21 December 2025

Abstract:

The accurate prediction of CO₂ capture in porous liquids is vital for optimizing the design and operation of carbon capture technologies. Several factors, including pressure, temperature, porous liquid type, and concentration (wt%), influence the CO₂ absorption process in porous liquids. However, despite the growing use of machine learning (ML) in modeling CO₂ solubility in ionic liquids and deep eutectic solvents, its application to porous liquids, a class of materials with unique structural and adsorption characteristics, remains largely unexplored. This study addresses this critical gap by developing and benchmarking a comprehensive suite of ML models to predict CO₂ capture efficiency in porous liquids. Advanced algorithms including CNNs, ANNs, SVMs, Random Forests, and Gradient Boosting Machines were trained and validated on 300 data points. Among these, CNN and XGBoost achieved the highest predictive performance, with R² values of 0.999 and 0.992 respectively. The MCOD algorithm confirmed dataset integrity, and sensitivity analysis revealed strong correlations between input parameters and CO₂ capture. SHAP analysis further identified pressure and porous liquid concentration as the most influential features. These insights can be practically applied to guide the formulation of porous liquid compositions, tune operating conditions (e.g., pressure and temperature), and prioritize experimental efforts toward high-impact variables, thereby accelerating the design of more efficient carbon capture systems. Additionally, the modeling framework can be extended to other gas–liquid systems, supporting future research in predictive environmental technologies.

Keywords: CO2 capture; Porous liquids; Data-driven models; Machine learning algorithms; Monte Carlo outlier detection; Sensitivity analysis

Full paper is available at www.springerlink.com.

DOI: 10.1007/s11696-025-04614-3

Chemical Papers 80 (5) 5131–5162 (2026)