Data-driven modeling of refractive index in LiBr/H2O solutions using machine learning: applications in optical interferometry

This research explores the refractive index behavior of LiBr/H2O solutions, emphasizing their significance in optical interferometry applications. A range of advanced machine learning techniques, including Multi-Layer Perceptron Artificial Neural Networks (MLP-ANN), Decision Tree (DT), Adaptive Boosting (AdaBoost), Ensemble Learning (EL), Least Squares Support Vector Machine (LSSVM), and Random Forest (RF), were utilized to model and predict the refractive index with high accuracy. The model’s effectiveness was improved by tuning hyperparameters with the Coupled Simulated Annealing (CSA) approach. A comprehensive dataset of 2240 experimentally derived data points was assembled, of which 1792 were used for training purposes and 448 were reserved for testing. The study incorporates key input variables including LiBr mass fraction (kg/kg, ranging from 0 to 0.619), temperature (°C, ranging from 23 to 47), and wavelength (nm, ranging from 436.1 to 655) which critically influence the refractive index. Sensitivity analysis, performed using Monte Carlo simulations, identified the LiBr mass fraction as the most significant factor (sensitivity value: 5.14), followed by temperature (3.32) and wavelength (1.57). The dataset was rigorously validated prior to modeling to ensure its suitability for predictive tasks. Performance evaluation demonstrated that the LSSVM model exhibited superior predictive accuracy, achieving the highest R2 values (0.999912293 for training, 0.999911005 for testing), the lowest RMSE (0.000603247 for training, 0.000616433 for testing), and the lowest AARE% (0.033524846 for training, 0.03501588 for testing). These findings underscore the effectiveness of machine learning, particularly LSSVM, in precisely predicting the refractive index of LiBr/H2O solutions. This approach offers a dependable, economical, and highly efficient substitute for conventional laboratory methods, allowing scientists to predict refractive index values with increased accuracy while requiring only minimal experimentation.