Abstract: Droplet drift generated during pesticide application by plant-protection Unmanned Aerial Vehicles (UAVs) has become a critical issue restricting the safe and sustainable development of UAV-based spraying technologies. Excessive spray drift not only reduces pesticide utilization efficiency but also induces a series of environmental and ecological risks, including environmental contamination, off-target deposition, and phytotoxic effects on non-target organisms. With the rapid expansion of UAV applications in agricultural production, especially in China, accurately characterizing and predicting droplet drift behavior under diverse operational conditions has become an urgent scientific and regulatory demand. To address these challenges, international standards such as ISO 22866:2005 have established field measurement protocols for spray drift, providing a standardized basis for quantitative drift assessment. However, due to the unique aerodynamic characteristics of multi-rotor UAVs—such as strong downwash airflow, low-altitude operation, and flexible flight parameters—traditional drift prediction models originally developed for manned aircraft or ground-based sprayers often exhibit limited applicability to UAV spraying scenarios. Therefore, it is necessary to develop UAV-oriented, data-driven prediction models that explicitly incorporate UAV operational features and environmental variability. In this study, based on the ISO 22866:2005 standard, a comprehensive dataset comprising 150 spray drift measurements was collected from four representative domestic plant-protection UAVs under typical operating conditions. To effectively capture the complex coupling relationships between multiple influencing factors and droplet drift rate, Spearman correlation analysis and Random Forest-based feature importance ranking were employed to identify key variables. On this basis, a droplet drift prediction model for plant-protection UAVs was constructed using a Backpropagation (BP) neural network, and SHAP (SHapley Additive exPlanations) was introduced to interpret the model outputs and quantify the contribution of each influencing factor. Furthermore, a UAV-specific droplet drift prediction software platform, UAVDP (UAV Droplet Drift Prediction Platform), was developed to facilitate practical application of the proposed model and was systematically compared with the widely used physical drift prediction tool AGDISPpro. The results demonstrate that the BP neural network model achieves high prediction accuracy and generalization capability, outperforming several conventional machine learning methods and showing significant accuracy improvements over AGDISPpro. SHAP-based interpretability analysis reveals that wind speed and spray flow rate are dominant positive contributors to drift, while operational height and nozzle number exhibit notable suppressive effects. Based on the model predictions, the drift risk level of plant-protection UAV operations was quantitatively assessed. Under compliant wind speed conditions (<5 m/s), the required downwind buffer distance is generally less than 20 m, and the associated drift risk lies between that of manned agricultural aircraft and large ground-based boom sprayers. Overall, this study provides a data-driven modeling framework, interpretable insights into drift mechanisms, and a practical prediction platform, offering robust technical support for precision pesticide application, drift risk control, and the formulation of safe UAV spraying standards.