Ensemble machine learning models for ternary electrolyte viscosity prediction

Accurate viscosity prediction in ternary electrolyte systems is essential for oil and gas drilling, geothermal operations, and brine processing. This comprehensive study compared three boosting-based ensemble machine learning algorithms - Gradient Boosting Regressor (GB), Extra Gradient Boosting (XGBoost), and Adaptive Boosting (AdaBoost) for predicting viscosity in Ca C l 2 -NaCl– H 2 O (Ternary 1, n=252) and KCl–NaCl– H 2 O (Ternary 2, n=122) systems using rigorous mathematical methodology comprising 92+ equations, 5-fold cross-validation generating 2,535 runs, and optimization across 507 hyperparameter configurations. Gradient Boosting achieved superior performance on both systems with R ² = 0.9946 (%AAD = 1.8746%) for Ternary 1 and R ² = 0.9852 (%AAD = 1.4280%) for Ternary 2, placing predictions within the “Excellent” industry standard, while demonstrating exceptional robustness by maintaining R ² > 0.98 despite 48% dataset reduction and achieving near-machine-precision convergence (error = 1.7 × 10 − 10 at 200 iterations) with minimal overfitting. XGBoost exhibited system-dependent performance with competitive results on complex divalent systems ( R ² = 0.9922) but degraded performance on simpler monovalent systems ( R ² = 0.9706, %AAD = 2.7686%) due to over-regularization, while AdaBoost proved unsuitable with %AAD > 3.7% reflecting fundamental misalignment between exponential loss and regression objectives. These findings establish Gradient Boosting as the optimal algorithm for ternary electrolyte viscosity prediction with geometric convergence rate providing theoretical justification for superiority, suitable for immediate industrial deployment, with future research directions including adaptive regularization strategies, extension to multicomponent systems, pressure effect incorporation, and exploration of deep learning alternatives.