Machine learning based calculation of refractive index of polyethylene glycol polymer

This study develops advanced machine learning algorithms to accurately predict the refractive index of polyethylene glycol (PEG) polymers using temperature and molecular weight as key input variables. A databank comprising 99 experimental measurements were utilized to train and assess multiple algorithms, including Extra Trees, Random Forests, Decision Trees, Ensemble Learning, AdaBoost, K-Nearest Neighbors, Multilayer Perceptron Artificial Neural Networks (MLP-ANN), Linear Regression, and Elastic Net. The dataset demonstrated a strong correlation between molecular weight and refractive index ( r = 0.64), and an inverse correlation with temperature ( r = −0.72). Model performance was assessed using R 2 , MSE, and AARE%, revealing that MLP-ANN, Linear Regression, and Elastic Net models yielded the highest accuracy on the testing dataset. Furthermore, SHAP analysis identified molecular weight as the dominant factor influencing refractive index predictions. The proposed approach significantly reduces the need for experimental measurements and offers a reliable, time-efficient solution for estimating the optical behavior of PEG polymers under varying conditions. • Predicting refractive index of polyethylene glycol using machine learning models. • Use of k-fold cross validation technique. • Outlier detection using Hat method. • Sensitivity analysis using SHAP method.