With the acceleration of urbanization, anthropogenic activities have induced pronounced modifications of surface morphology, posing severe challenges to the sustainable development of historic cities. This study aims to characterize the spatiotemporal distribution of surface deformation in the old urban area of Xi'an and to elucidate its driving factors, and further to develop a single-feature prediction model based on cumulative subsidence to reveal deformation trends and spatial disparities over the coming year. Using Sentinel-1A imagery acquired between 2019 and 2024, we applied PS-InSAR and SBAS-InSAR techniques to extract time-series deformation and construct a comprehensive subsidence dataset, cross-validating the correlation and reliability of the two InSAR results through a dual-technology mutual-checking approach. Long-term deformation sequences derived from PS-InSAR were then employed to train and validate two forecasting frameworks—N-BEATS and LSTM. The results indicate: (1)The deformation fields retrieved by PS-InSAR and SBAS-InSAR exhibit a high degree of linear correlation and display a "northwest subsidence and southeast uplift" spatial pattern, with a maximum annual subsidence rate of 9.283 mm/a and a maximum cumulative subsidence of 61.16 mm; and a maximum annual uplift rate of 5.585 mm/a and a maximum cumulative uplift of 49.727 mm. (2) Comparative evaluation of the forecasting models demonstrates that N-BEATS, with superior single-feature learning capability, predictive accuracy, and stability, outperforms LSTM. Forecasts from the N-BEATS model project a maximum cumulative subsidence of 74 mm and a maximum cumulative uplift of 47.466 mm over the next year, with an amplified contrast between northwest and southeast deformation.