As a worldwide environmental and geological disaster, land subsidence may cause serious harm to urban development. Therefore, the prediction of land subsidence is a key scientific problem. Decheng county, Shandong Province in China is taken as the research object. Based on BP neural network (BPNN) and random forest (RF) method, the analysis and prediction of regional land subsidence are carried out by applying multi-source monitoring data, Geographic Information System (GIS), and machine learning algorithm. Combined with Short Baseline Synthetic Aperture Interferometric Radar (SBAS-InSAR) and GIS technology, the spatiotemporal evolution characteristics of land subsidence from 2017 to 2020 are analyzed. The impact of different groundwater levels on land subsidence is quantitatively analyzed by BPNN and RF algorithm. The real-time prediction model of regional land subsidence is established. The results show that: 1) The area with the most serious land subsidence is located in Songguantun town, the maximum annual average subsidence rate is −40.71 mm/yr. 2) Land subsidence is mainly affected by deep groundwater and shallow groundwater in the research area. 3) The accuracy of the prediction results of the BPNN model is higher than that of the RF model when groundwater level change is used to predict land subsidence.