<p>This work proposes an accurate quantitative structure–property relationship (QSPR) model for glass transition temperatures (<i>T</i>_gs) of 315 polymers with diverse structures. Support vector regression (SVR), combined with a genetic algorithm, was used to develop the SVR model that achieved a determination coefficient (<i>R</i>²) of 0.937 and a root-mean-square (<i>rms</i>) error of 25.047 K for 79 <i>T</i>_gs of polymers in the test set. The investigation shows that increasing the percentage of C atoms, molecular polarity, complementary information content index, spectral mean absolute deviation and eigenvalue in augmented adjacency matrix by introducing benzene, naphthalene, anthracene, pyridine, quinoline, imide, and/or C‒N atom pairs with a topological distance of 9, can result in high chain rigidity and high <i>T</i>_gs. Conversely, increasing the free volume and flexible segments by introducing groups such as ‒Si‒O‒, ‒Si‒C‒, and ‒Si‒N‒ in the backbone chain, N and O atom pairs, S and X (heteroatoms) atom pairs with a topological distance of 1, F‒X with a topological distance of 2 and CH₂RX groups can bring down <i>T</i>_gs.</p> Graphical Abstract <p></p>