This diploma thesis focuses on the use of machine learning
techniques to discover relationships between biological
communities and the environmental conditions in which they
live. For this, data from the monitoring of bird communities of
windbreaks in southwestern Slovakia were used.

Various machine learning methods including random forest,
artificial neural network, support vector machine and Bayesian
classifier were compared with linear and logistic regression
models using selected metrics. These methods were used to model
bird species richness, Shannon's biodiversity index of bird
communities and the presence of three indicator species: common
blackbird (Turdus merula), great tit (Parus major) and common
pheasant (Phasianus colchicus) in the windbreaks of
southwestern Slovakia. The interpretation of these models was
performed using permutation feature importance, SHAP feature
importance and accumulated local effects plots. Consistent with
the literature, a positive relationship between the diversity
of the bird community and the width of the windbreak was found,
and the most important features in modeling the presence of
indicator species were also consistent with species'
characteristics. The machine learning methods were found to be,
with one exception, more effective than traditionally used
models, with artificial neural network and support vector
machine models being the most suitable. However, artificial
neural network may be slightly preferable. The thesis also
pointed out the importance of developing new tools to help with
interpretation of results, especially for machine learning
methods that are otherwise entirely uninterpretable.