Application of machine learning models for predicting zinc oxide nanoparticle size

Accurate characterization of nanostructures is essential for optimizing their properties for various applications, yet conventional methods such as electron microscopy are costly and time-consuming. This study explores the potential of machine learning (ML) in predicting the size of zinc oxide (ZnO) nanoparticles using synthesis conditions and band gap. A dataset of 90 samples, comprising nine synthesis parameters, was compiled from published literature. These samples were divided into training (75 %) and testing (25 %) sets, and four ML models—Catboost (CB), Gradient Boosting (GB), Extreme Gradient Boosting (XGB), and a Stacking Ensemble—were implemented, with hyperparameter tuning performed using Randomized Search CV. Among these models, the Stacking Ensemble approach achieved the highest accuracy, with an R² value of 0.9377 and a mean absolute error (MAE) of 3.08 nm. Feature importance analysis identified the band gap as the most significant predictor of nanoparticle size, followed by calcination temperature, reaction time, precursor concentration, and reaction temperature. To further validate the model, an additional set of 25 unseen experimental datasets from previous studies was used, where the model closely predicted 17 instances (68 %). Additionally, ZnO nanoparticles were synthesized, and their size was estimated at 53.07 nm by the ML model, closely aligning with the scanning electron microscopy (SEM)-measured size of 58.9 nm. These findings underscore the potential of ML as a cost-effective alternative to traditional size characterization techniques. To enhance practical application, a user-friendly graphical interface (GUI) was developed, providing a scalable solution for nanoparticle size estimation while reducing reliance on experimental characterization and accelerating materials research.