Climate change is leading to an increase in the frequency and intensity of flooding, making it necessary to consider future changes in flood risk management. In regions where ground-based observations are significantly restricted, the implementation of conventional risk assessment methodologies is always challenging. This study proposes an integrated remote sensing and machine learning approach for flood risk assessment in data-scarce regions. We extracted the historical inundation frequency using Sentinel-1 SAR and Landsat imagery from 2001 to 2023 and predicted flood susceptibility and inundation frequency using XGBoost, Random Forest (RF), and LightGBM models. The risk assessment framework systematically integrates hazard components (flood susceptibility and inundation frequency) with vulnerability factors (population, GDP, and land use) in two SSP-RCP scenarios. The results indicate that in the SSP2-RCP4.5 and SSP5-RCP8.5 scenarios, combined high- and very-high-flood-risk areas in the Ili River Basin in China (IRBC) are projected to reach 29.1% and 29.7% of the basin by 2050, respectively. In the short term, the contribution of inundation frequency to risk is predominant, while vulnerability factors, particularly population, contribute increasingly in the long term. This study demonstrates that integrating open geospatial data with machine learning enables actionable flood risk assessment, quantitatively supporting climate-resilient planning.