This study investigates the quantitative structure-permittivity relationship (QSRR) of dielectric constants (ε) for a diverse set of polymers using a machine learning approach. A transparent model was developed by combining genetic algorithm (GA) and multiple linear regression analysis (MLRA) to create a mechanistically explainable model. The dataset consists of 71 polymers with various structures, including polyvinyls, polyethylenes, polyoxides, polystyrenes, and others. The molecular descriptors were calculated based on the structure of the repeating monomer units, and 523 descriptors were selected after filtering. Two models, one with four variables and the other with eight variables, were proposed. The best model showed high predictive performance with R² values of 0.905 and 0.812 for training and test sets, respectively. The models were validated using various methods, including leave-one-out cross-validation, y-scrambling, and external validation. The descriptors involved in the models include mean atomic Sanderson electronegativity, mean information index on atomic composition, R maximal autocorrelation, and others. The study concludes that the developed models can effectively predict dielectric constants based on polymer structure, which can be useful for designing new polymers with desired electrical properties.This study investigates the quantitative structure-permittivity relationship (QSRR) of dielectric constants (ε) for a diverse set of polymers using a machine learning approach. A transparent model was developed by combining genetic algorithm (GA) and multiple linear regression analysis (MLRA) to create a mechanistically explainable model. The dataset consists of 71 polymers with various structures, including polyvinyls, polyethylenes, polyoxides, polystyrenes, and others. The molecular descriptors were calculated based on the structure of the repeating monomer units, and 523 descriptors were selected after filtering. Two models, one with four variables and the other with eight variables, were proposed. The best model showed high predictive performance with R² values of 0.905 and 0.812 for training and test sets, respectively. The models were validated using various methods, including leave-one-out cross-validation, y-scrambling, and external validation. The descriptors involved in the models include mean atomic Sanderson electronegativity, mean information index on atomic composition, R maximal autocorrelation, and others. The study concludes that the developed models can effectively predict dielectric constants based on polymer structure, which can be useful for designing new polymers with desired electrical properties.