<p>Slope erosion hazards induced by rainfall scouring extremely threaten civilian safety, such as landslides and subgrade subsidence, especially in the southern part of China. Nevertheless, there are deficiencies in the maintenance and slope erosion protection comments. In this paper, model tests were conducted to analyze the key erosion characteristics of influencing factors on subgrade slopes, including bare slopes (BS1 ~ BS4, representing different rainfall intensities, slope gradients, and grain size distributions), slopes protected by the sprayed substrate (SS), barbed wire reinforcement (BWR), and vegetation coverage (VC) under multiple working conditions. Under the influence of rainfall scouring, the saturation of shallow soil rapidly increased, leading to a reduction in permeability and the formation of ponding, which impeded rainwater infiltration and gave rise to runoff, scouring, and erosion. A higher rainfall intensity (RI) and slope gradient (SG) or loss of small particles could decrease the anti-erosion performance (AEP). Protective measures significantly enhanced the AEP. Finally, an integrated evaluation framework was established by synergizing model tests with an ensemble machine learning model (XGBoost-LightGBM-CatBoost), which was trained on 10,000 data sets generated from validated physical models. Verified through ten case studies in Guangdong, China, this framework demonstrates high accuracy in assessing slope erosion risk (A-D levels) and provides precise, cost-effective protection recommendations (e.g., optimizing plant fiber content or vegetation coverage over costly BWR), thereby filling the existing research gap in scouring maintenance and offering valuable guidance for subsequent slope construction design.</p>