Soil salinization poses a significant ecological and environmental threat to sustainable development in arid and semi-arid regions worldwide. Existing research predominantly focused on experimental plots or individual oases, lacking a systematic understanding of soil salinization across the entire oasis region in arid Northwest China. To clarify the spatiotemporal patterns and key driving factors of soil salinization in this region and to support precise ecological restoration and water–soil resource management. In this study, we developed a soil salinity estimation model to characterize the spatiotemporal distribution of saline soil in oasis regions of arid Northwest China from 2019 to 2023. The model adopted the Tree-structured Parzen Estimator (TPE) algorithm within the Optuna Bayesian hyperparameter optimization framework to optimize XGBoost parameters, integrating multi-source remote sensing data, in-situ soil salinity measurements, and environmental covariates, achieved satisfactory predictive performance (RPD = 1.41). In addition, the optimal geographical detector model was employed to identify key driving factors and their interactions influencing salinization patterns. The results indicated that: (1) From 2019 to 2023, the saline soil area in this region exhibited a spatial pattern of high in the south and low in the north. Slightly saline soils accounted for the largest proportion (>59%), mainly distributed in the Northern Xinjiang, Ili, and Alxa-Hexi Corridor oases. In contrast, highly saline soils were significantly concentrated in the Southern Xinjiang oases, where the proportion of saline soil area exceeded that in the Ili oasis by 65%. (2) During 2019-2023, the total area of saline soils in the oases of the arid region of Northwest China decreased by 2,892.43 km2, with the overall spatial pattern remaining relatively stable (variation about 3%), with the Ili oasis being the only region showing a slight increase (+26.20 km2). Salinization showed a fluctuating pattern of “increase–decrease–increase,” with a rebound observed in some local areas during 2022-2023. Salinization severity generally improved, as evidence by a shift from high to low salinization levels. Specifically, 42.36% of severely saline soils converted to moderately saline, and 53.63% of extremely severe saline soils converted to severely saline, indicating the most pronounced improvements. (3) Based on the Geographical Detector model calibrated with optimal parameters (Optimal-Parameter Geographical Detector, OPGD), we found that groundwater level was identified as the dominant factor affecting salinization spatial heterogeneity (q=0.315), followed by potential evapotranspiration (q=0.295) and precipitation (q=0.291). All factor interactions exhibited nonlinear enhancement effects, with the interactions between groundwater level ∩ land use and groundwater level ∩ precipitation showing the strongest explanatory power (q≥0.43).