Machine Learning Method (Decision Tree) to Predict the Physicochemical Properties of Premium Lebanese Kishk Based on Its Hedonic Properties

Abstract

This study sets the criteria of high-grade kishk (a dried fermented cereal–milk product) based on sensory attributes. For this, kishk samples were collected, and physicochemical attributes and sensory attributes were recorded. Subsequently, Spearman’s correlation between sensory properties and physicochemical properties was calculated. A decision tree [DT] was applied with the mean total sensory score [MTSC] as the dependent factor to establish the physicochemical factor/s upon which the different kishk grades were set. To compare the physiochemical attributes of the different grades, the general linear model was applied. Moisture content is negatively and significantly correlated with most sensory attributes. Titratable acidity [TA] is positively and significantly correlated with most sensory attributes. The DT analysis showed that TA was the classifying factor [p = 0.01], and accordingly, grade A [TA ≥ 4.56], grade B [2.50 p = 0.018], the protein-to-fat ratio [P:F] [p = 0.027] and pH [p < 0.001] differ significantly between the different kishk grades. Accordingly, the criteria for grade A kishk are TA ≥ 4.56, pH ≤ 3.95, moisture < 4%, P:F < 2.03, and particle density < 1489. The low pH and moisture content render it a shelf-stable high-acid food.

Keywords

• machine learning
• decision tree
• kishk
• grades
• physicochemical properties
• sensory attributes