The Ordos region, as a typical ecologically fragile area in the arid and semi-arid northern region of China, plays an important role in the construction of ecological civilization. Understanding the vegetation changes and their driving impacts in this region is of great significance. Considering the non-linear relationship between driving factors and vegetation, this study, based on the kNDVI index that can more accurately characterize vegetation state, first analyzes the spatiotemporal evolution characteristics of vegetation in the Ordos region from 1990 to 2023. Then, the non-linear functional relationship between driving factors and vegetation is quantified using the machine learning model XGBoost and the interpretable model SHAP. The results show that: (1) the kNDVI in the Ordos region shows a distinct upward trend, with a change slope of 0.001; the areas with significant vegetation increase are mainly concentrated in the eastern part. (2) The contribution rates of palmer drought severity index (PDSI), vapor pressure deficit (VPD), precipitation (PRE), and temperature (TEM) to vegetation change are 34.31%, 28.10%, and 23.69%, respectively. (3) The impact of driving factors on vegetation is not a simple linear relationship. Specifically, PDSI and VPD show a quadratic polynomial relationship with kNDVI, and PRE and TEM show a cubic polynomial relationship with kNDVI. The research results help reveal the impact mechanism of climate change on fragile ecosystems.