Modeling sound speed of aqueous polyethylene glycol solutions

Polyethylene glycol (PEG) water mixtures are used in many scientific and industrial processes, where their acoustic behavior is closely tied to formulation design and performance. In this work, we develop a modeling approach driven by data to estimate the speed of sound in aqueous PEG solutions over a wide range of operating conditions. A databank of 395 experimental records, including temperature, polymer weight molality, and PEG molar mass as inputs, was assembled from the literature and screened using a Monte Carlo–based outlier detection procedure to ensure internal consistency. Several machine learning regressors including Support Vector Regression (SVR), Decision Tree, AdaBoost, Random Forest, K-Nearest Neighbors (KNN), Ensemble Learning, and a Multilayer Perceptron artificial neural network (MLP-ANN) were trained and evaluated using multiple statistical indicators and graphical analyses. Correlation analysis and relevance factors show that polymer weight molality is the dominant descriptor of sound velocity (correlation ≈ 0.91), while temperature (≈ 0.17) and PEG molar mass (≈ 0.09) play secondary roles. Among the tested algorithms, SVR provides the most reliable generalization, attaining the maximum coefficient of determination and the minimum error measures on the test set, whereas Decision Tree and KNN exhibit clear signs of overfitting. Global sensitivity assessment and SHAP-based interpretation confirm the leading influence of polymer weight molality, followed by temperature and PEG molar mass. The suggested structure offers a fast and economical alternative to purely experimental approaches for predicting sound velocity in PEG solutions, and can support the design and optimization of PEG-based systems in chemical, materials, and process engineering.