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

Estimating solubility of nitrogen in various ionic liquids: application of machine-learning

Kassem Al Attabi, Farzona Alimova, Anupam Yadav, H. S. Shreenidhi, Abhinav Kumar, Devendra Singh, Vatsal Jain, Zainab Ahmed Hamodi, Aseel Smerat, and Ahmad Khalid

Department of Computers Techniques Engineering, College of Technical Engineering, The Islamic University, Najaf, Iraq

E-mail: kassem.alattabi@iunajaf.edu.iq

Received: 15 November 2025  Accepted: 26 December 2025

Abstract:

The limited solubility of nitrogen (N₂) in conventional solvents such as water severely restricts the efficiency of ammonia synthesis, motivating the search for alternative media. Ionic liquids (ILs), with their tunable physicochemical properties, offer a promising solution, yet the vast chemical diversity of cations and anions makes experimental screening highly complex. This study aims to develop robust machine-learning models capable of accurately predicting N₂ solubility in diverse ILs using COSMO-based descriptors. A comprehensive set of algorithms, including Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Random Forests, CatBoost, Support Vector Regression (SVR), and others underwent training and validation with a dataset comprising 385 points in total, with performance assessed via numerous metrics and visual analyses. The results demonstrate that ANN, CNN, Random Forest, CatBoost, and SVR achieved superior predictive performance, with SHAP analysis identifying pressure and specific descriptors (e.g., S5) as dominant factors influencing solubility. The novelty of this work lies in integrating advanced ML models with interpretable SHAP-based analysis, providing both predictive accuracy and mechanistic insights into N₂–IL interactions. While challenges remain due to the limited size of available datasets and the immense diversity of IL structures, this study highlights the potential of data-driven approaches to accelerate IL design. Future perspectives include expanding experimental databases and incorporating hybrid physics-informed ML frameworks to further enhance generalizability and guide rational IL development for energy and environmental applications.

Keywords: Energy technologies; N2 solubility; Machine learning; Ionic liquids; Data-driven models

Full paper is available at www.springerlink.com.

DOI: 10.1007/s11696-025-04623-2

Chemical Papers 80 (4) 3419–3440 (2026)