On the application of Maxent algorithms in ecology

The article discusses the logical and computational foundations of the maximum entropy method, which the MaxEnt program uses. This program makes it possible to build models of the placement of different species of animals and plants. The subject of the analysis is the method of maximum entropy as a criterion for the success of the selection of model parameters. The principles of its work are shown in a series of increasingly complex quantitative examples from ecology. The calculations are illustrated by programs in the R language, which can be performed by readers for a deep understanding of the meaning of the procedure. The emphasis is placed on the difference between MaxEnt technology and other classifiers (discriminatory analysis, neural networks, etc.): instead of using contrast between groups of objects, MaxEnt strives to capture and enhance the uniformity of objects of the same group. This almost automatically leads to the separation of objects of one studied status from another. This technique makes it possible to effectively perform classification constructions in conditions of information scarcity. Some approaches for assigning a "break point", a threshold for binary classification, including elements of ROC analysis, the use of percentiles and quantiles are considered. The article serves as a practical introduction to the technology of constructing classifications using the principle of maximum entropy.