Deriving Real Estate Meta Data from CityGML LOD 2 Models

In the German purchase price database, millions of transactions are recorded in just few years. The information is necessary to assess the local real estate markets. To evaluate the purchase prices correctly it is necessary, that all information that has an impact on the price is included in the evaluation process. Information about the inside of the real estate can only be obtained by questionnaires or construction files until now. However, the response rate of questionnaires and access to construction files are limited. Unfortunately, for that reason, many data-gaps exist in the samples. The goal of this approach is to derive the information about the inside geometry and fill these data-gaps in the database. With the German CityGML level of detail 2 (LOD2) building models it is easier to derive the characteristics about the size. Every building in the model has information about height, surroundings, shape of the roof and geographical location.In this approach, we use a parametric and a non-parametric method to derive the size of the living space, the number of floors besides the existence of a basement and the usable attic from LOD2 models. As training and validation data, we use several thousand building models and the existing information about the inside geometry from the purchase price database which was derived by questionnaires and construction files. In the literature we can find information that shows, that the size of the walls and utility shafts depend on the building age. We therefore expect that the usage of the information about the building age as an influencing parameter can improved the accuracy of the estimation.We use a cross-validation approach with parametric and non-parametric regression to derive the information about the inside geometry from the outside geometry and compare the methods.We expect that non-parametric methods over-perform parametric methods. To use the approach in the future for the missing data, the premise that the missing data in the database is missing at random has to be true.In near future, further data (digital twin – LOD3+) can be used to derive the information even more accurate.