A Machine Learning Approach for Classifying Rheological Behavior and Stability of Polymer Solutions Using a Naive Bayes Classifier

The rheological characterization of polymer solutions is critical for optimizing their performance in various industrial applications, including enhanced oil recovery, food processing, and biomedical engineering. Traditional experimental methods for classifying flow behavior, viscosity, shear-thinning degree, and system stability are often time-consuming and require expert interpretation. This study presents a data-driven approach to automate these classifications using a Gaussian Naive Bayes (GNB) classifier. A comprehensive dataset comprising 420 experimental measurements of polymer solutions, sourced from Kaggle, was used. The dataset includes key physicochemical features: shear rate, polymer concentration, NaCl concentration, Ca²⁺ concentration, temperature, and logarithmic viscosity. These features were used to predict four distinct target labels: Newtonian vs. Non-Newtonian flow behavior, Shear Thinning Degree, Viscosity Class, and System Stability. The GNB model, chosen for its suitability with continuous features and probabilistic foundation, was trained and evaluated using an 80/20 train-test split and 5-fold cross-validation. The model achieved high test accuracies of 0.857 (Flow Behavior), 0.798 (Shear Thinning Degree), 1.000 (System Stability), and 0.940 (Viscosity Class). Cross-validation scores confirmed model robustness. Permutation feature importance analysis revealed that the slope of the log(viscosity) vs. log(shear rate) relationship, temperature, and Ca²⁺ concentration were the most influential features across all classification tasks. This study demonstrates the efficacy of simple yet interpretable machine learning models in accurately predicting complex rheological properties from basic formulation parameters, paving the way for rapid, low-cost material screening and process optimization.