Machine learning and geospatial technologies

Machine learning (ML) has recently been recognised as a promising generalisation technique and even called by some researchers a new paradigm for cartographic generalisation. The research described in this chapter follows this paradigm shift in the map generalisation process as it aims to design and verify ML models for automatic settlement selection. First, we outline the concept of ML as well as ML types and selected ML models. Second, we present a case study of automatic settlement selection for small-scale maps using ML models. We apply deep learning (DL), random forest (RF), decision tree (DT), and decision tree supported with genetic algorithms (DT_GA) models to a varied settlement data sample. Using ML models, we automatically classify settlements as selected and omitted. To evaluate the ML models, we validate the results against the selection status acquired from an atlas reference map, and then compare the performance across the different ML models. The obtained selection accuracy for each tested model, understood to be the similarity to the settlement selection on atlas reference map, was very high, ranging from 78% (DT) to nearly 84% (RF).