Application of Splitting (Cross Validation) Technique and Herzberg Equation in Feed Pellet Quality Prediction Models

The present study had the objectives to apply the splitting technique in feed pellet quality prediction models and validate these models through Herzberg equation. The independent factors (input data) were the particle size (PS: coarse and medium), heat treatment (HT: expanded-pelleted and pelleted), fat addition levels (FA: 15, 25, 35 and 45 g/kg of feed) and moisture addition levels (MA: 0, 7, 14 and 21 g/kg of feed) which were combined in a full factorial design (2 × 2 × 4 × 4), resulting in 64 different treatments with eight replicates each. The intact pellets amount and pellet durability index (PDI) were considered as the model’s output data. In the splitting technique the whole data set, composed by 512 observations, were splitted in two data set: (1) model construction set(75% of the total data) and (2) model validation set (25% of the total data set, which were selected randomly from the original data set). Both equations, the one obtained by splitting method and the one obtained by whole data set, had good coefficient of determination and similar residues square means. It was concluded that splitting technique can be successfully applied to fit a prediction equation for feed pelleting process and that Herzberg equation consists in a useful tool to validate the coefficient of determination of those models.