CLASSIFICATION TECHNIQUE OF MACHINE LEARNING AS SPECIES DISTRIBUTION MODEL FOR EXOTIC FISH IN RIVERS

Abstract

The prevalence of exotic species has been a major ecological problem all over the world, and Japan is no exception. This study modelled distributions of five exotic fish species (channel catfish, bluegill, largemouth bass, smallmouth bass and mosquitofish) in major rivers in Kanto region by using classification techniques: classification and regression trees (CART) and random forest (RF). National Census on River Environments was used as a response and predictor variables. Both models showed high prediction accuracy for all exotic species, and RF outperformed CART. Suitable habitat ranges of the fish estimated by RF were well accorded with the ranges reported based on observations. Additionally, the result was presented also as the species distribution map as an example of its application. Overall, this research demonstrated the importance of the combination between advanced statistical approach (CART and RF) and detailed environmental data.