High-frequency oscillatory anomalies of radon concentration in Diangou spring in Wudu City, Gansu Province often clustered after the Wenchuan earthquake. The radon observation data and seismic catalogue were preprocessed by using the data mining method, then the parameters such as radon anomaly, magnitude, focal depth, and seismic zonation were extracted. On this basis, Bayesian network model was adopted to study the statistical relationship between radon anomaly in Diangou spring and seismic activity in the surrounding area. The results of seismic zonation obtained from the clustering method were basically consistent with those of tectonic division, reflecting the clustering characteristics of seismic activities in different structural units. The results of Bayesian network model showed that the gas radon anomalies were related to the surrounding seismicity, especially the seismic activities in the middle segment of Longmenshan fault and the Qingchuan fault. The research results are helpful to understand the differential characteristics of gas radon response in Diangou spring to surrounding earthquakes, thus having reference significance for the judgment of gas radon anomaly and earthquake prediction in the future. It also provides a new technical idea and method for earthquake precursor anomaly analysis and seismic activity zonation.