With the large-scale exploitation of mineral resources, surface subsidence in mining areas has become an increasingly serious issue, posing significant threats to environmental safety and the sustainable development of these regions. Taking Ping'an No. 1 Mine and Shuangyang Phosphate Mine in Yangshui Mining Area, Kaiyang County, Guizhou Province as research objects. time-sequence synthetic aperture radar interferometry (InSAR) technology was integrated with machine learning algorithms to monitor and predict surface subsidence in this phosphorus mining area. The small baseline subset technique was employed to obtain cumulative time series data of surface subsidence from September 2020 to May 2023. Additionally, global positioning system(GPS) data were used to validate the accuracy of the monitoring results. Important subsidence points E1 and E2 in the study area were selected to establish autoregressive integrated moving average (ARIMA) and dung beetle optimization (DBO)-optimized long short-term memory (LSTM) prediction models. The ARIMA-DBO-LSTM combination model, using different weights, was employed to predict and analyze settlement trends. Results indicate that the prediction accuracy of the ARIMA-DBO-LSTM combination model, based on the weight assignment of the residual inverse method, is the highest at both E1 and E2 points. This combination model compensates for the limitations of individual prediction models when addressing complex time series data. The proposed method provides valuable insights for predicting surface subsidence in the study area.