Progressive Multi-target Optimization with Quality Gates (PMTO-QG) Using Machine Learning Classifier for Formulation Optimization and Physical Quality of Honey Powder

This study aimed to develop a predictive system for optimizing honey powder formulation through a Progressive Multi-Target Optimization with Quality Gates approach integrated with machine learning classifiers. The research was conducted using experimental dataset of honey powder production, including moisture content, HMF, bulk density, particle density, true density, solubility, and flowability. Three algorithms will be compared to see which is the best, namely Random Forest, Lasso Regression, and XGBoost used to classify and predict the best formulation. Quality gates were established as layered checkpoints to ensure each predicted formulation met the required standards before advancing to subsequent stages. Results from stage I analysis demonstrated that the PMTO-QG framework effectively filtered suboptimal formulations while improving prediction efficiency and accuracy compared to conventional trial-and-error methods. The system successfully identified formulations parameters within acceptable ranges, providing a robust foundation for subsequent experimental validation. The predicted formulation will be validated through physical tests including yield, particle size distribution, microstructure, color attributes, Tg temperature, stability tests, and sensory testing of powdered honey. This approach highlights the potential of integrating data-driven modeling and quality assurance checkpoints in functional food product development.