SOLVING OF CLASSIFICATION PROBLEM IN SPATIAL ANALYSIS APPLYING THE TECHNOLOGY OF GRADIENT BOOSTING CATBOOST

In the paper two models of spatial analysis are considered. The models are dedicated for spatial analysis of ecological factors distribution, such as distribution of contaminant concentration on researched territory. The models are created using the method of machine learning - gradient boosting. In order to build the models we have used open source effective library CatBoost. Functions AUC and Accuracy were calculated for each model. MultiClass - integrated function of CatBoost library was used for loss minimization. For solving the problem, it was necessary to define affiliation of searched point from test dataset to one of four classes. This problem belongs to the type of classification, or rather multiclassification. Asa result of the studies, an effective model was obtained that allows one to perform with sufficient accuracy the spatial forecast of the factor distribution at points and regions of the studied field with an unknown gradient value of this factor. This model works adequately with a training dataset of 0.5% of all analyzed information about the object.