SHAP analysis of electrospinning parameters for predicting nanoparticle size, using a gradient-boosted model to predict particle diameter and to explain which process variables drive it.
Electrospray is a common method for fabricating PLGA microparticles and nanoparticles used in drug delivery, but the relationship between process parameters and the resulting particle size is notoriously non-linear. This project predicts the mean diameter of electrosprayed particles from a pre-trained XGBoost model, then runs SHAP (SHapley Additive exPlanations) analysis to make the model interpretable. It shows the importance and interaction of each parameter rather than treating the model as a black box.
The target is mean particle diameter in micrometers. SHAP plots show which variables matter most and how each one pushes a given prediction higher or lower, the kind of insight a process engineer needs to dial in a formulation.
Particle size governs release kinetics, bioavailability, and how a drug-delivery particle behaves in the body. An interpretable predictor turns expensive trial and error at the bench into a guided search, and the SHAP layer lets researchers reason about the model's recommendations. The work builds on and credits the dataset and models from Wang, F. et al., Machine learning predicts electrospray particle size (Materials & Design, 2022).