Predicting shrimp growth: Artificial neural network versus nonlinear regression models

This study evaluated the potential of artificial neural networks (ANN) as an alternative to the traditional statistical regression techniques for the purpose of predicting shrimp growth in a commercial setting. Empirical data collected from a commercial shrimp farm in Hawaii was used for this evaluation. Eight regression functional forms (i.e., linear, polynomial, log reciprocal, von Bertalanffy, Gompertz, logistic and exponential) were employed as counterparts to ANN. The specific models were first estimated from a dataset consisting of 459 records and then applied to a dataset consisting of 249 records to validate their predictive performance. Performance was assessed by four measures (i.e., root mean square error (RMSE), R2, the percentage of wrong turning points and the percentage of predicted values that are within the 5% tolerance of the corresponding actual values). The results indicated that ANN outperformed regression models for the complex set of conditions typical of a commercial production environment. ANN generated a slightly better descriptive shrimp growth curve than the best ones generated from nonlinear models and made the most accurate prediction. Hence, we conclude that ANN represents a valuable tool for predicting shrimp growth for the variable set of growth conditions typifying commercial shrimp farms.