Accurate prediction of water activity in ionic liquid-based aqueous ternary solutions using advanced explainable artificial intelligence frameworks

The accurate determination of water activity in ionic liquid-based aqueous mixtures is crucial for various scientific and industrial applications. Water activity is a critical parameter that affects the stability, shelf-life, and reactivity of aqueous systems and solvents. Numerous machine learning studies have focused on predicting this thermodynamic property, but the majority have concentrated on binary mixtures of ionic liquids, while the modelling of ternary mixtures remains relatively underexplored. This study addresses this research gap by successfully integrating machine learning and evolutionary algorithms to accurately estimate water activity in ionic liquid-based ternary aqueous solutions. Specifically, the study employed Categorical gradient boosting technique optimized with three evolutionary algorithms: Grey Wolf optimization, Whale optimization, and Gravity search optimization algorithms. A comprehensive dataset of 1,829 experimentally measured values from the literature were utilized to develop and validate the proposed models. Among them, the CatBoost-Grey Wolf Optimization model exhibited the best performance, achieving remarkable accuracy with an R² of 0.993, an Average Absolute Percentage Error of 0.13%, and a Root Mean Squared Error of 0.00265. To ensure the credibility of the dataset and the stability of the predictive framework, the leverage approach was applied. Furthermore, model interpretability was enhanced using Shapley Additive Explanations, allowing a clearer understanding of the impact of input variables. Trend analysis further verified the model’s ability to capture key physical relationships. Ultimately, this study highlights the model’s utility as a powerful and efficient tool for optimizing ionic liquid-based ternary aqueous systems, significantly reducing reliance on extensive experimental measurements.