As a crucial parameter in estimating precipitable water vapor from tropospheric delay, the weighted mean temperature (Tm) plays an important role in Global Navigation Satellite System (GNSS)-based water vapor monitoring techniques. However, the rigorous calculation of Tm requires vertical profiles of temperature and water vapor pressure that are difficult to acquire in practice. As a result, empirical models are widely used but have limited accuracy. In this study, we use three machine learning methods, i.e., random forest (RF), backpropagation neural network (BPNN), and generalized regression neural network (GRNN), to improve the estimation of empirical Tm in China. The basic idea is to use the high-quality radiosonde observations estimated Tm to calibrate and optimize the empirical Tm through machine learning methods. Validating results show that the three machine learning methods improve the Tm accuracy by 37.2%, 32.6%, and 34.9% compared with the global pressure and temperature model 3 (GPT3). In addition to the overall accuracy improvement, the proposed methods also mitigate the accuracy variations in space and time, guaranteeing evenly high accuracy. This study provides a new idea to estimate Tm, which could potentially contribute to the GNSS meteorology.