This study investigates quantitative structure-property relationships for polymers using machine learning techniques, leveraging the ATHAS data bank. Developed by Wunderlich et al., this data bank contains specific properties of polymers. These properties encompass essential information, including the glass transition temperature (Tg0) of fully amorphous polymers, the heat capacity difference at Tg0, the equilibrium melting temperature, the heat of melting point at 100 % crystallinity, Θ temperature, the number of skeletal vibrations, and the temperature dependence of heat capacities, etc. Our methodology focuses on establishing correlations between these polymer-specific physical properties and the repeating structural units within the polymers. To achieve this, we employed various regression models, including artificial neural networks with equivalent and damping structures, Lasso, Random Forest, and XGBoost regression. Model validations were conducted using 5-Fold cross-validation. Our results indicate that the artificial neural network model exhibited the lowest error. Furthermore, we showcase the practical application of quantitative structure-property relationships in predicting the properties of two polymers not yet included in the ATHAS data bank.