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

Estimation of lithium adsorption in brine via robust machine learning algorithms

Farag M. A. Altalbawy, Ahmad Almalkawi, Anupam Yadav, H. S. Shreenidhi, Vishnu Saini, Farzona Alimova, Devendra Singh, Vatsal Jain, Ahmad Alkhayyat, and Ahmad Khalid

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

E-mail: ahmedalkhayyat85@gmail.com

Received: 26 October 2025  Accepted: 21 January 2026

Abstract:

Accurately forecasting lithium adsorption in brine under different conditions is crucial for advancing energy technologies. This research focused on developing robust machine learning models to predict lithium adsorption, considering factors like brine composition, operational conditions, and adsorbent properties. We tested numerous machine learning techniques including XGBoost, CatBoost, CNNs, and SVRs on a 599-point dataset, confirming its suitability with Monte Carlo outlier detection. Among all models, XGBoost achieved an R² of 0.94 with RMSE = 0.021, while CatBoost reached an R² of 0.93 with RMSE = 0.024, outperforming other approaches. Sensitivity analysis revealed that lithium concentration, adsorbent surface area, and temperature were the most critical parameters, contributing over 65% of the variance in adsorption outcomes according to SHAP analysis. These conclusions underline the power of advanced approaches, especially XGBoost and CatBoost, for predicting lithium adsorption, offering valuable insights for industry and future efficiency improvements.

Keywords: Lithium adsorption; Energy technologies; Machine learning; Adsorbent; Data-driven models

Full paper is available at www.springerlink.com.

DOI: 10.1007/s11696-026-04693-w

Chemical Papers 80 (4) 4297–4322 (2026)