Predicting shrimp disease occurrence: artificial neural networks vs. logistic regression

Predicting the occurrence of disease outbreaks in aquacultural farms can be of considerable value to the long-term sustainable development of the industry. Prior research on disease prediction has essentially depended upon traditional statistical models with varying degrees of prediction accuracy. Furthermore, the application of these models in sustainable aquaculture development and in controlling environmental deterioration has been very limited. In an attempt to look for a more reliable model, we developed a probabilistic neural network (PNN) to predict shrimp disease outbreaks in Vietnam using farm-level data from 480 Vietnamese shrimp farms, including 86 semi-intensive and 394 extensive farms. We also compared predictive performance of the PNN against the more traditional logistic regression approach on the same data set. Disease occurrence (a 0-1 variable) is hypothesized to be affected by a set of nearly 70 variables including site characteristics, farming systems, and farm practices. Results show that the PNN model has a better predictive power than the logistic regression model. However, the PNN model uses significantly more input (explanatory) variables than the logistic regression. The logistic regression is estimated using a stepwise procedure starting with the same input variables as in PNN model. Adapting the same input variables found in the logistic regression model to the PNN model yields results no better than the logistic regression model. More importantly, the key factors for prediction in the PNN model are difficult to interpret, suggesting besides prediction accuracy, model interpretation is an important issue for further investigation. (C) 2000 Elsevier Science B.V. All rights reserved.