Parasite infracommunities as predictors of harvest location of bogue (Boops boops L.): a pilot study using statistical classifiers

Abstract

The accuracy of classifying bogue (Boops boops) according to the fishery from which it was harvested was evaluated by applying several statistical classification techniques to fish parasite abundances. Bogue captured in 2001 in two fisheries off the Atlantic coast of Spain were compared with one off the Spanish Mediterranean coast. One hundred bogue were classified to each harvest location (fishery) using different numbers of parasite species chosen as predictors by a best subset method. Two parametric methods of classification (linear and quadratic discriminant analysis) were compared with two non-parametric approaches (k-nearest neighbour classification and feed-forward neural network) and the cross-validated correct classification rate determined in each case. The best results were achieved for k-nearest neighbour classification with 96% of fish being correctly assigned to their fishery. That result was based on five predictor parasite species, namely Aphanurus stossichii, Bacciger israelensis, Hemiurus communis, Microcotyle erythrini and Lecithocladium excisum. The optimal classification for a feed-forward neural network was very similar to that achieved with linear discriminant analysis at 94% correct classification rate for only four predictor species. Quadratic discriminant analysis performed worst of the four classification methods examined. Classification accuracies for all four statistical approaches were, however, remarkably similar and were quite accurate with only between 4 and 8% of fish incorrectly classified regardless of the method used. With greater temporal and spatial resolution of sampling effort, this technique holds promise as a cost-effective method of distinguishing harvest locations of bogue that can readily be adapted to other fish species.