A genetic Takagi-Sugeno fuzzy system for fish habitat preference modelling

Genetic fuzzy systems have a potential to be applied to ecological studies as a tool for species distribution modelling and habitat evaluation. However, no study has focused on how different model formulations affect habitat preference evaluation and performance of the model. The present study therefore aims to assess the effect of model formulations on habitat preference evaluation through the optimization process. We employed a genetic algorithm (GA)-optimized Takagi-Sugeno fuzzy model for evaluating habitat preference of topmouth gudgeon (Pseudorasbora parva), a freshwater fish in Japan. The model was trained based on the mean square error (MSE) between composite habitat preference and observed presence-absence, and evaluated using confusion matrix-derived performance measures such as kappa and correctly classified instances (CCI). The present results clearly illustrated the effect of model formulations on habitat preference evaluation, which appeared as different trends in habitat preference curves (HPCs) and the variance. The use of the product equation is recommended in view of model accuracy and consistency in HPCs. Further studies would be necessary for better understanding of model behaviour to different conditions of data such as sample size and prevalence.