A novel fuzzy based approach for inducing diatom habitat models and discovering diatom indicating properties

For a given physico-chemical parameter, diatoms have specific optimum and tolerance, thus they can be used as indicators. Many diatoms in the relevant ecological literature can be found as indicators for a certain water quality (WQ) or trophic state index (TSI) class, but for many other diatoms the indicating properties are unidentified. Noticeable progress in this direction has been made by classical decision trees. However, these methods have several drawbacks in diatom classification. To overcome this problem, in this paper we use a novel fuzzy approach for discovering the diatoms' indicating properties. The prediction accuracy of the diatom habitat models (DHMs) depends on the types of fuzzy operators, the similarity metric, and the number and shape of the membership functions (MFs). We use several MFs to describe the relationship between the diatoms and the abiotic factors. Two novel similarity metrics are introduced in order to increase the models' classification accuracy. We compare the description/prediction accuracy of our approach and several classical and fuzzy based classification algorithms. Also, we compare the proposed similarity metrics with an existing similarity metric used for inducing DHMs. The evaluation results show that the proposed approach has higher prediction accuracy than the other methods, and is less prone to over-fitting when the novel similarity metrics are used. The obtained models are verified with the known ecological references, and they were also used for adding new ecological references for some diatoms. Thus, the induced DHMs could serve as basis for establishing a new fresh-water indicator system.