In the presented research, the main aim is the assessment of machine learning (ML) techniques usage in the process of acquiring and formalizing generalization rules and variables, understood as settlement features, in settlement generalization. The research specifically addresses the problem of automated settlement selection for 1:1,000,000 scale. We focus on two processes of cartographic knowledge formalization: first, the extraction of semantic and structural knowledge through data enrichment, and second, automated acquisition and application of procedural cartographic knowledge with the use of ML models. This work contributes to extending the toolbox for small‐scale mapping and we concentrate specifically on decision tree‐based models. The main achievements of our research are as follows. First, we verify the existing settlement’s selection rules and variables proposed for 1:1,000,000 detail level by comparing the selection results with the well‐elaborated reference map. Second, with the use of DT‐based models, we make the part of the cartographic knowledge hidden in maps explicit, and show how these models can be used to explore and formulate additional generalization rules and variables important in settlement selection. The research is carried out for the area of Poland but we believe it can be validated and extended to other National Mapping Agencies.