Figure 2
The S H A P summary plot shows a horizontal axis labeled “S H A P value (impact on model output)” ranging from negative 50 to 250 in increments of 50 units and a vertical axis listing fifteen product attributes. From top to bottom, the attributes are labeled “Calories,” “Price,” “Packaging size,” “Mean price of price promotions,” “Mean number of price promotions,” “Plain flavor (yes or no),” “Sugar content,” “Protein content,” “Fat content,” “Brand (yes or no),” “Fruit flavor (yes or no),” “Production method (organic yes or no),” “Other flavor (yes or no),” “Packaging material (glass yes or no),” and “Low fat option (yes or no)”. Each attribute line contains clusters of colored dots distributed horizontally. The dots range in color from purple to yellow, corresponding to low and high values of the analyzed product characteristics, as shown by the color scale bar on the right labeled “Value of analyzed product characteristic,” with “Low” at the bottom in purple and “High” at the top in yellow. The data for each product attribute is shown as follows: Attribute: Calories; Range: negative 50 to 0, S H A P Value: Medium, Range: 0 to 120; S H A P Value: Low. Attribute: Price; Range: negative 50 to 0, S H A P Value: Medium, Range: 0 to 70, S H A P Value: Low. Attribute: Packaging size; Range: negative 40 to 0, S H A P Value: High, Range: 0 to 10, S H A P Value: Low. Range: 10 to 40, S H A P Value: High. Attribute: Mean price of price promotions; Range: negative 10 to 10, S H A P Value: High, Range: 10 to 40, S H A P Value: Medium, Range: 40 to 100, S H A P Value: Low. Attribute: Mean number of price promotions; Range: negative 30 to 0, S H A P Value: Low, Range: 0 to 20, S H A P Value: Medium, Range: 20 to 90, S H A P Value: High. Attribute: Plain flavor (yes or no); Range: negative 20 to 0, S H A P Value: Low, Range: 20 to 90, S H A P Value: High. Attribute: Sugar content; Range: negative 50 to 10, S H A P Value: Medium, Range: 10 to 120, S H A P Value: Low. Attribute: Protein content; Range: negative 40 to 80, S H A P Value: Medium. Attribute: Fat content; Range: negative 60 to negative 20, S H A P Value: Low, Range: negative 20 to 70, S H A P Value: Medium. Attribute: Brand (yes or no); Range: negative 20 to 0, S H A P Value: High, Range: 0 to 10, S H A P Value: Low. Range: 10 to 40, S H A P Value: High. Attribute: Fruit flavor (yes or no); Range: negative 10 to 0, S H A P Value: Low, Range: 0 to 10, S H A P Value: High. Range: 10 to 30, S H A P Value: Low. Attribute: Production method (organic yes or no); Range: negative 30 to 0, S H A P Value: High, Range: 0 to 10, S H A P Value: Low. Attribute: Other flavor (yes or no); Range: negative 20 to 0, S H A P Value: High, Range: at 0, S H A P Value: Low. Attribute: Packaging material (glass yes or no); Range: negative 15 to 0, S H A P Value: High, Range: 0 to 10, S H A P Value: Low. Attribute: Low fat option (yes or no); Range: negative 10 to 0, S H A P Value: High, Range: at 0, S H A P Value: Low, Range: 0 to 10, S H A P Value: High.Bee swarm plot of the SHAP analysis. Notes: This graph presents individual data points without overlap, creating a “swarming” effect. It reveals the distribution, density and variation of data along a numeric variable more effectively than scatter or box plots. In SHAP analysis, it summarizes how top attributes impact the model's output. Each dot represents an instance on each attribute row. The horizontal axis shows SHAP values, while the vertical axis orders product attributes by their mean absolute SHAP values. Yellow indicates high, and purple indicates low attribute values. For categorical attributes, high represents “yes” and low represents “no” (Lundberg and Lee, 2017). Source: Authors’ own work