In semiconductor physics, precise ZnO nanoparticle size determination is crucial due to its impact on material properties and applications. Traditional methods like electron microscopy were limited by their complexity and high costs, posing challenges for researchers with limited resources. Machine learning offered a promising alternative for nanoparticles size prediction. This study utilized supervised machine learning to develop an efficient method for determining ZnO nanoparticle size. Models are trained with dataset of 149 samples collected from previous studies on ZnO nanoparticle synthesis and characterization, encompassing nine synthesis parameters. These samples were split into training (75%) and testing (25%) sets. Four tree based machine learning models—Random Forest (RF), CatBoost Regressor (CatBoost), Gradient Boosting (GB), and Extreme Gradient Boosting (XGBoost)—were employed. We optimized their performance by tuning hyperparameters manually and evaluated using Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²). Among the models, XGBoost effectively captured relevant patterns in the data, with R², RMSE and MAE of 0.6875, 4.6634 and 3.6713 respectively, leading to superior predictive performance. Energy band gap was found to have the highest impact on the particle size followed by reaction temperature and calcination hour. The model is then validated with use of 27 additional datasets of unseen experiments from previous studies. Despite its moderate R², the model significantly closely predicted 21 instances (78%). The findings of this study illustrate the potential of machine learning in improving nanoparticle size determination, providing a more efficient and cost-effective alternative to traditional methods.