ML-augmented topological indices for predicting glycol physicochemical properties

This study establishes a robust quantitative structure-property relationship (QSPR) framework for glycols by integrating symmetry-aware topological indices, including Zagreb and neighborhood-degree descriptors, with machine learning (ML) models such as support vector regression (SVR) and random forest (RF). We demonstrate that these indices exhibit strong positive correlations R^2>0.90 for all topological indices with key properties including molecular weight (MW), boiling point (BP), heat capacity (HC), and flash point (FP), validated through both linear/quadratic regression ( p < 0.05) and ML models. Quadratic regression outperformed linear approaches, achieving 27.4% higher accuracy for all indices prediction, while SVR captured non-linear relationships with FP, R^2=0.9988 . Graphical validation confirmed deviations of less than 1% from experimental data, enabling reliable property estimation without lab trials. This work provides a computationally efficient alternative to experimental characterization and a template for green chemistry applications through reduced synthetic iterations, and ML-augmented indices for future small-molecule studies.