Hierarchical ZnMn2O4 microspheres for selective ethylene glycol sensing: synthesis optimization via machine learning

Abstract

This study not only demonstrates the potential of engineered hierarchical ZnMn₂O₄ (ZM) nanostructures as high-performance toxic volatile organic compounds (VOCs) sensors, but also proposes a machine-learning framework to predict and optimize the specific surface area (SSA)—a key factor in solid–gas sensing performance—for metal-oxide semiconductors by varying pivotal synthesis parameters, including hydrothermal temperature, calcination temperature, and solution pH (adjusted via acid, base, or surfactant additives). A full factorial design yielded 180 distinct synthesis combinations, which were used to train Random Forest, Gradient Boosting, and XGBoost regression models. XGBoost outperformed other models (R² = 0.92, RMSE = 2.68 m²/g, RRMSE = 8 %, and MAE = 1.95 m²/g) and was employed to predict the optimal synthesis conditions. Specifically, the model predicted a maximum SSA of 78.65 m² g⁻¹ at a hydrothermal temperature of 167.14 °C, a calcination temperature of 440.82 °C, and pH = 9.14; an experimental SSA of ∼74 m² g⁻¹ was obtained under these settings (≈8 % deviation). Experimental validation using two new samples with modified pH confirmed prediction errors below 9 %. The optimized ZM3 sample exhibited a crystallite size of ∼18 nm, increased lattice strain, and hierarchical nanoplatelet morphology. Gas sensing tests revealed that ZM3 showed the highest response (3.69–500 ppm ethylene glycol) at 185 °C, together with rapid response/recovery times of 12 s and 14 s, and excellent selectivity over other VOCs. This study presents a reproducible and data-driven methodology for synthesis optimization and microstructural control in spinel oxides. The full Python scripts and supplementary characterization results are provided in Appendix A to ensure transparency and facilitate reproducibility.