**3. Results and discussion**

#### **3.1 Database construction and cleaning**

The data obtained from literature was standardized as follows. All components for the formulation (oil content, egg yolk, water content, flavorings, and vinegar) were recalculated to be expressed as percentage (%). Viscosity was standardized in Pa�s. All sensory scores were found in different scales (e.g.: 1–5, 1–9, 1–10, 1–100), therefore, the data was rescaled to a range of 1–10. The rescaling was done multiplying the score by 10 and dividing the result by the higher value of the original scale.

For selecting the appropriate features of the database, a correlation matrix was generated (**Figure 2**). It was observed that few characteristics (taste and texture scores) were closely correlated (�0.75) with the acceptability; meanwhile, the oil content, viscosity, and egg yolk content showed positive correlation but with lower impact on the desired variable (0.00 — 0.25). In the opposite way, the water, vinegar, and stabilizers content were negatively correlated with the acceptability in a range of �0.50 to �0.25. The features with better correlation (oil content, viscosity, taste score, and texture score) were selected for the acceptability prediction.

It is well known that all ingredients in the mayonnaises play a specific and impor-
