An integrated data-driven and decision tree framework for modeling quality attributes of infrared-dried apple slices

This study developed a novel integrated quality modeling framework that combines multivariate data analytics with an interpretable decision-tree approach to predict and classify quality attributes of apple slices processed by intermittent infrared drying. Physicochemical, structural, and sensory variables, including effective moisture diffusivity, hardness, shrinkage, rehydration capacity, reduced density, browning index, polyphenol oxidase inactivation time, vitamin C retention, and consumer acceptance, were evaluated across nine drying treatments varying slice thicknesses (5 - 13 mm), temperatures (70 – 80°C), and drying durations until reaching a constant weight. Correlation analysis revealed strong process-quality relationships: thicker slices increased diffusivity and shrinkage but reduced vitamin C retention and acceptance, while higher temperatures intensified hardness and browning. Principal Component Analysis (PCA) explained 93% of the total variance, distinguishing a nutrient–texture axis (62.34%) from a functional–consumer axis (30.67%). Principal Component Regression (PCR) enabled accurate linear prediction ( R² > 0.9, MSE < 0.2). Hierarchical clustering and multidimensional scaling (MDS) identified three distinct treatment groups, with the treatment of (thick = 13 mm, 80°C, time = 255 min) exhibiting the lowest acceptance among all treatments. Decision-tree analysis provided interpretable process thresholds, showing that vitamin C retention dropped below 0.65 when thickness > 7 mm and temperature ≥ 75°C. Thus, thin slices dried at 70 – 75°C achieved a sustainable balance among drying efficiency, nutrient retention, and consumer acceptance. These findings provide a foundation for developing real-time quality monitoring strategies, offering practical guidance on intelligent optimization tools for industrial infrared drying applications.