Predicting blood–polymer interactions in electrospun nanofiber systems using interpretable AI models: A meta-analytical approach

Advanced biomaterials are essential for restoring tissue function, yet ensuring the safety of blood-contacting devices remains a significant challenge. This study presents a data-driven framework to predict the hemocompatibility of electrospun nanocomposites by correlating their nanoscale physicochemical signatures with biological outcomes. Using a curated dataset of 99 experimental samples, we evaluated the predictive power of support vector machines (SVM), linear discriminant analysis (LDA), and multilayer perceptron (MLP) architectures. To address data scarcity and class imbalance, the synthetic minority oversampling technique (SMOTE) was integrated into a 10-fold cross-validation pipeline. Results indicate that the MLP model provides superior predictive performance, achieving a validation accuracy of 98.4%. Furthermore, the application of explainable AI (XAI) methods to elucidate feature contributions revealed that surface roughness, contact angle, and thermal degradation onset are primary determinants of blood compatibility, potentially eliminating the need for redundant coagulation assays. This research establishes a cost-effective, non-invasive methodology for the high-throughput screening of next-generation medical device coatings.