Exploring Data Mining Algorithms for Predicting Duck Egg Weight Based on Egg Quality Characteristics

Abstract

The present investigation aimed to compare the performance of twomachine learning algorithms, Artificial Neural Network (ANN), and Classification and Regression Tree (CART), alongside the Automatic Linear Modelling (ALM), and the traditional Multivariate Linear Regression model (MLR) to predict the egg weight (EWT) of Mallard duck from some egg traits including egg length (EL), egg width (EWd), egg shape index (ESI), eggshell weight (ESW), albumen weight (AW), albumen height (AH), yolk weight (YW), yolk height (YH), yolk diameter (YD), and Haugh unit (HU). The Pearson correlation between observed and predicted values (r), coefficient of determination (R 2 ), adjusted coefficient of determination (R 2 adj), Root Mean Squared Error (RMSE), and Relative Approximation Error (RAE) were used to estimate model performance. EWT hada strong correlation with egg dimensions (EL and EWd, r=0.752 and 0.790, respectively), AW (r= 0.815), and YW (r= 0.784). The R 2 adj values were 0.981, 0.970, 0.964 and 0.897, for ANN, ALM, MLR, and CARTmodels, respectively. The lowest RMSE was found for ANN (0.753), while the highest RMSE was observed for CART (1.778). Overall, the ensemble models proposed in this study yielded similar results, with the ANN algorithm showing a marginally superior predictive performance compared to ALM, CART, and MLR models. This finding suggests that ANN could be considered the most suitable for the prediction of egg weight in Mallard duck.