With the widespread application of water-saving irrigation, the distribution and accumulation pattern of soil salinity have changed significantly. However, little is known about changes in soil salinity profiles in inland river basins under this context. Here, we analyze the evolutionary characteristics of soil salinization in the Sangong River Basin in the arid region of northwestern China, based on soil profile salinity data from 2015 and 2023. We found an overall increase in soil salinity during this period, with the accumulation pattern shifting from the bottom to the middle layers of the soil profile. Compared with other machine learning models, the integrated model (XGBoost + LightGBM) showed superior predictive performance, achieving an R-2 of 0.72 and an RMSE of 4.16 g/kg. From 2015-2023, soil salinization intensified across the basin profile, with the proportion of severely salinized areas increasing significantly (by similar to 17 %). The main factors influencing soil salinity change include land surface temperature, temperature, and soil organic carbon. Identification of salinization risk areas indicated that the proportion of at-risk cropland and barren land increased in 2023 compared with 2015. Soil salinity accumulation in the profile has intensified under water-saving irrigation, and now poses a potential threat to agricultural productivity and ecological security in drylands. Our findings provide valuable insights for developing land management policies and salinity mitigation strategies.