Abstract: Landslide disasters occur frequently in the mountainous regions of western China, and accurate classification of landslide hazard risk levels is crucial for effective geohazard mitigation. Data-driven models have achieved notable progress in landslide susceptibility evaluation; however, challenges persist at large-scale, small-area levels due to limited dynamic features and excessive dependence on sample purity. To address this, this paper proposes a new integrated evaluation method based on the fusion of time-series InSAR deformation data with machine learning models. In this approach, InSAR-derived deformation is introduced as a dynamic factor, and multicollinearity analysis is employed to screen and optimize evaluation indicators. Subsequently, various machine learning methods are used to construct a landslide susceptibility evaluation model, obtain the optimal susceptibility prediction model for large-scale and small-area landslides, and then carry out the landslide susceptibility evaluation while accounting for InSAR-based surface deformation and multi-temporal landslide catalogues. In this paper, Heifangtai in Gansu Province is selected as the study area, and the results show that the Maxent model has the best accuracy in the large-scale, small-area conditions. In addition, compared with models that exclude InSAR data, the incorporation of time-series deformation features improves the model's prediction accuracy by approximately 0.36%, and increases the proportion of landslides identified within very high and high susceptibility zones by 4.29%. Using an eight-year time landslide catalogued data as positive samples, the model achieves an accuracy of 99.26%, highlighting the effectiveness of long time-series, high-quality data in improving the accuracy and reliability of regional susceptibility assessments. This study contributes a dynamic evaluation framework that enhances the accuracy of landslide susceptibility mapping and offers valuable insights for risk assessment in large-scale, small-area geohazard-prone regions.