An Application of One-vs-One Method in Automated Taxa Identification of Macroinvertebrates

Abstract

Freshwater ecosystems face numerous anthropogenic stressors. For solving long-term effects in aquatic ecosystems due to the human-induced actions, we need to use benthic macro invertebrates instead of a chemical analysis. The use of benthic macro invertebrates requires their identification which is a laborius and cost-intensive task. By means of automated taxa identification of macro invertebrates the costs can be reduced and the identification process can be speeded up. However, the identification demands reliable tools. In this research we extended the use of one-vs-one method from Support Vector Machines into several other classification methods and we examined the tie situation problem which is encountered in one-vs-one method. Overall, we used 15 different classification methods in this paper. By thorough experimental tests we achieved 96.8% accuracy by using Support Vector Machines with the quadratic kernel. Tie situation analysis revealed that ties were more frequent when using Support Vector Machines together with one-vs-one classification framework and majority voting method than other classification methods.