Multi-output machine learning prediction of drying kinetics and quality attributes during hot-air drying of banana slices

Abstract This study develops an integrated experimental–computational framework to model drying kinetics and quality evolution of banana slices during hot-air drying. Experiments were conducted on unripe and ripe samples under controlled temperature conditions, with moisture ratio, water activity, and hardness measured over time. A multi-output machine learning approach was employed using Random Forest, XGBoost, and Artificial Neural Network models to simultaneously predict drying behavior and quality attributes. The results show that the proposed models achieve high predictive accuracy, with XGBoost providing the most consistent performance ( R 2 = 0.9775 and RMSE = 0.0492 for moisture ratio prediction). The findings highlight the role of moisture ratio as a key state variable linking drying kinetics to quality evolution, indicating that changes in water activity and texture are governed by moisture removal. This study provides a concise and interpretable framework for predicting and controlling drying processes in food systems.