Café’s Performance Modeling with Spatial Data

The relevance of the article lies in the importance of the placement problem for the economic performance of organizations and the growth of interest in the use of spatial data in decision support systems in recent years. The main purpose of the research work is to model the estimation of impact of important spatial features for café’s turnover prediction. The article analyzes some approaches that combine spatial data with machine learning to solve the placement problem. A correlation analysis of spatial data has been carried out. A multistage feature selection for two sets of features proper for different types of models was made. The hyperparameter optimization for the selected modeling methods (linear regression, decision tree, random forest, gradient boosting) was made and models were created. The main tools are the Python programming language and its libraries pandas, sklearn, XGBoost, hyperopt, shap, boostaroota. The analysis of the obtained results was carried out. The gradient boosting model was identified as optimal in terms of accuracy and interpretation. The result of the work is the created approach to modeling the economic performance of a company using machine learning based on spatial data.