Data-Driven Machine Learning Approach for Modeling the Production and Predicting the Characteristics of Aligned Electrospun Nanofibers

Abstract

The generation of electrospun nanofibrous with controlled size, shape, and spatial orientation is crucial for the development of biomedical and electronic devices. Aligned nanofibers are advantageous over random nanofibers because control of the spatial orientation can improve electrical and optical properties and play an important role in tissue engineering applications, impacting the mechanical and biological properties of the scaffold. Therefore, different machine learning models have been developed to predict the optimal production of electrospun-aligned poly(vinyl alcohol) nanofibers. The database was obtained by multiple assays using the airgap electrospinning setup and varying the voltage, the distance between the tip and collector, and polymer concentration. Binary classification models were developed, which can predict the production or not of aligned nanofibers. In addition, regression models have been developed to predict the orientation, angle, and diameter of the nanofibers when there is a production of nanofibers. A convolutional neural network has also been developed. It was concluded that for the binary classification, the artificial neural network performs better predictions obtaining an accuracy equal to 0.94, and for the validation set, an accuracy equal to 0.90 and an F1-score equal to 0.87 were obtained.