This dataset is created as part of work package WP3 of the Deep Geo Stat project, as part of the ESS topic B5674-2020-GEOS (project 101033951: 2020-NL-GEOS-DEEP-GEO-STAT). The objective was to research whether a Siamese Convolutional Neural Network (SCNN) could be used to automatically detect changes in land use. At Statistics Netherlands, every few years, the so-called "Bestand Bodemgebruik" (BBG) is created, which is a file containing information about the land use of the Netherlands for a given year. The country is split up into many polygons, whereby each polygon is labelled with the most common type of land use for that specific area. Creating the BBG is a time-consuming process and the assumption is that it can be speed up if we can automatically detect the changes in land use. During the research, SCCNs were created to decide the changes in land use for three classes. The SCCN models were then applied to the whole of The Netherlands. The results were then converted to GIS files. This dataset contains the GIS files for the classes 34 (building sites), 51 (other agricultural), and 60 (forest). These files can be loaded in QGIS, for example. Each polygon contains the prediction as it rolls out of the model to which we add the label (0: changed, 1: unchanged). Predictions are made on polygons that belonged to the specific class in 2017 and the new label is the prediction for 2018 and 2020.