This study utilizes mobile signal data as an emerging data source, aiming to construct a framework that can efficiently distinguish homecoming visitors from other types of tourists. The rich behavioral features contained in mobile signal data are deeply analyzed. A two-stage clustering method based on machine learning, which integrates multiple aspects such as feature engineering, model construction, and performance evaluation is designed to ensure the accuracy and generalization ability of classification results. The main data source is the signaling data of the three major telecommunications operators in China. Using the signaling data from China's three major telecommunications carriers as the primary data source, this study verified the number of tourists in the Guangxi, as well as in Nanning, Guilin, and Yulin cities during the 2019 National Day Golden Week. The results indicate that the method presented in this paper offers a viable new approach for the statistical analysis and data mining of large-scale tourism figures.