Running is most common for daily exercise in modern societies. However, there are a limited number of studies on the thermal comfort experienced by people while they run. Moreover, the predicted mean vote (PMV) model, is restricted in terms of the accurate prediction of dynamic change with the thermal environment, such as that finite experiment when dealing with the body movement situation. In order to remedy this shortcoming, we designed a series of experiments, including data collection by questionnaire survey, data processing by Multilayer Perceptron model and machine learning logistic regression model, and finally successfully established a method to capture thermal comfort under dynamic conditions.