Application of unsupervised machine learning techniques to assessment of quality habitat

Abstract

Abstract As a result of human activities, river valleys have changed: river beds have been transformed as a result of their regulation and dam construction. This paper presents unsupervised machine learning techniques to distinguish arthropod communities and attempts to explain the ecological priorities of individual species based on them. Data on a group of 95 species of carabid beetles from 16 habitats on the floodplains of two rivers - Bytytsia and Strilka (Dnipro basin, Ukraine). Analyses were performed using the program R version 4.1.1. Fuzzy clustering was performed using the fanny function from the R cluster package, and visualization of the results was performed using the t-SNE method from the Rtsne package. In our analysis, the following habitat type characteristics were chosen to distinguish communities: closed (forest) or open (grassland). According to the results of fuzzy clustering, out of 95 carabid species, 37 species were selected whose probability of belonging to their cluster was at least 0.95. These species form distinct three groups. The first group of carabid species is associated with forest sites in Bytytsia. The second group is connected with occurrence on meadow sites in Strilka. The third group of species is connected mainly with the meadow sites of the Bytytsia River. Established groups of species reflect the current ecological situation on floodplains and the influence of human activities on it.