<p>Polyimides have broad applications such as aerospace, microelectronics, and fuel cells, where a high glass transition temperature (<i>T</i>_g) is critical for thermal stability and performance. Quantitative structure–property relationship (QSPR) offers a powerful approach to accelerate the discovery and design of polyimides by enabling <i>T</i>_g prediction from molecular structure. In this work, the dataset of 502 <i>T</i>_gs of polyimides was collected. Firstly, 208 descriptors were generated using the RDKit, from which the 9 most representative descriptors were selected through a series of screening methods. The random forest (RF), k-nearest neighbors (KNN), and extreme gradient boosting (XGboost) were used to build the QSPR models. Among the QSPR models studied, the XGboost-based model gave the most accurate results with the root mean square error (RMSE) of 4.35. It is extensively validated by internal and external validation strategies. The leave-one-out cross-validation (Q<Stack> _Loo ² </Stack>) of three<i> T</i>_g QSPR models was greater than 0.8. The external validation coefficient (Q<Stack> _Ext ² </Stack>) of three QSPR models exceeded 0.950. The mechanisms based on the developed models suggested that the parameters of the number of rotatable bonds, hydrogen bonds, heteroatoms, the length of molecular chains, and the mass of molecules will significantly influence the <i>T</i>_g of polyimides. This work provides reliable models for predicting the <i>T</i>_g of polyimides, which is expected to guide the rational design of novel polyimides with enhanced <i>T</i>_g.</p>