Nitric oxide absorption in green neoteric solvents: Robust models based on hybrid machine learning algorithms

Accurate determination of nitric oxide (NO) solubility within green neoteric solvents is vital for clarifying gas separation processes, optimizing industrial purification tracks, and advancing environmental engineering and green chemistry frameworks. In this research, advanced machine learning approaches were deployed to quantify NO absorption behavior based on diverse thermodynamic and structural inputs. The key parameters integrated into the predictive frameworks encompassed structural properties (density of hydrogen bond acceptors [HBAs], density of donors [HBDs], density of deep eutectic solvents [DESs], and number of moles of HBA), transport characteristics (viscosity of deep eutectic solvents [DESs]), and environmental operating conditions (temperature and pressure). Gradient Boosting Decision Tree (GBDT) architectures were developed and tested leveraging a comprehensive dataset containing 292 distinct experimental data points. To augment the predictive accuracy of the base GBDT network, four metaheuristic optimization techniques including Batch Bayesian Optimization (BBO), Evolution Strategies (ES), Bayesian Probability Improvement (BPI), and Gaussian Processes Optimization (GPO) were hybridized with the algorithm. The predictive validity of each configured pipeline was rigorously quantified using the coefficient of determination (R 2 ), Mean Squared Error (MSE), and Average Absolute Relative Error (AARE%). Cross-model comparisons demonstrated that the hybridized GBDT-BBO framework yielded superior predictive performance, securing an overall R 2 of 0.999365883 and a total MSE of 0.009283739. Global model interpretability achieved through SHAP (SHapley Additive exPlanations) values revealed that the density of hydrogen bond acceptors (HBAs) exerted the most dominant control over gas solubility levels, followed sequentially by DES density, temperature, pressure, DES viscosity, HBD density, and the number of moles of HBA. Concurrently, an inspection of algorithmic execution speeds indicated that the GBDT-GPO configuration exhibited unmatched computational efficiency, registering the shortest optimization runtime at 145 seconds. Ultimately, these outcomes validate the high utility of integrated hybrid machine learning methodologies for predicting complex gas solubility dynamics in sustainable media, providing an intelligent diagnostic instrument to evaluate neoteric fluid design and optimize industrial chemical modeling.

Hali tarjima qilinmagan