Machine Learning-Based Prediction of Temperature-Driven Solubility Changes in Aqueous Salt Solutions

This work explores the predictive capabilities of machine learning techniques in modeling the solubility behavior of KCl, BaCl₂, and NiCl₂ as a function of temperature. Employing Random Forest and Gradient Boosting algorithms, the study compared predicted solubility values with experimental data to assess model performance. Random Forest exhibited the highest reliability, especially for BaCl₂ and KCl, due to its ensemble approach and resistance to data irregularities. Although Gradient Boosting effectively captured nonlinear patterns, its accuracy declined at lower temperatures for NiCl₂ and KCl, likely due to unmodeled chemical factors such as complexation and ionic strength fluctuations. The findings affirm the relevance of machine learning for forecasting chloride solubility, highlighting NiCl₂ as a system sensitive to multifactorial interactions. Future improvements may involve integrating additional descriptors such as solution pH, total ionic strength, and ligand availability.