Abstract: In order to consider the influence of personalized factors on thermal comfort, a thermal comfort prediction model based on XGBoost classification algorithm was established. The unique thermal coding method was used to convert the original data into characteristic parameters, and the converted data was used as the input of XGBoost classification algorithm. After iterative training, the optimal thermal comfort prediction model of public buildings was obtained. The SHAP value was used to explain the characteristic parameters of the model, and the key factors affecting personalized thermal comfort were obtained. The results show that the area under the curve (AUC) of receiver operating characteristic (ROC) of thermal comfort prediction model based on XGBoost classification algorithm, are 0.95 and 89% respectively, which are better than those of the random forest algorithm models such as logistic regression, support vector machine, neural network, and showing high prediction accuracy. The key factors affecting personalized thermal comfort are air temperature, relative humidity, air speed and body weight.