XGBoost-based predictive modeling of droplet size and generation rate in single and double microfluidic emulsions

Microfluidic droplet generation has emerged as a powerful tool in life-science applications due to its ability to precisely control fluid dynamics at the microscale. However, accurately predicting droplet diameter and generation rate remains challenging because the underlying multiphase behavior arises from highly nonlinear interactions among flow conditions, device geometries, and fluid properties, particularly in systems involving single and double emulsions. In this study, we address this challenge by developing a predictive framework based on the XGBoost machine-learning algorithm to estimate droplet diameter and generation rate with high accuracy across a wide range of devices and operating conditions. The framework leverages a comprehensive experimental database and employs a two-stage modeling strategy in which droplet diameter is first predicted and subsequently used as an intermediate variable for estimating generation rate. Evaluation of the model demonstrates excellent agreement with experimental measurements across diverse geometries and flow regimes. Furthermore, SHAP-based feature importance analysis reveals that key geometric and physical parameters dominate droplet formation behavior, providing valuable insights for microfluidic system optimization. Additional analyses show that the proposed framework not only performs robustly under standard operating conditions but also maintains high reliability when exposed to significant variations in flow and fluid properties. This capability makes the model a practical and efficient tool for designing next-generation microfluidic devices, particularly for complex biomedical applications. Overall, the developed methodology can substantially accelerate the design, testing, and optimization processes of droplet-based microfluidic systems while reducing experimental cost and development time.