When modelling transport processes, the source data often contains geographical coordinates: latitude and longitude. But in intelligent geoinformation models (in particular, fuzzy networks and hypergraphs), the coordinates themselves are unsuitable features. Based on them, it is necessary to construct such features that will be later used to describe fuzzy and linguistic variables. The article proposes a method of such feature engineering and the specifics of its practical implementation. At the beginning, we consider the case when the dataset object contains one pair of coordinates. Here clustering methods are used. But on their basis, many additional ones can be introduced. Clusters contain hyperparameters, the selection of which for a specific application task can significantly improve model quality metrics. If a dataset object contains several pairs of coordinates, various metrics are numerical features: distances, bearings. The possibilities of feature engineering are further expanded by the interaction between departure and arrival clusters: distances between cluster centres, distances in the number of cells, statistical characteristics of the target between clusters, and many others. On rough terrain, it is important to consider not only latitude and longitude, but also altitude. With known latitude and longitude, the altitude can be loaded from external geographic information systems. Accordingly, we can calculate the height difference and the direction of descent or ascent. The article presents the implementation of the considered methods of feature construction in Python using the functionality of the sklearn package. Examples show how these introduced features change the quality of models in specific machine learning tasks.