Background and Challenge
Real-estate business is booming. In the year 2018 alone, roughly 300 000 real-estate deals were made by private individuals in Germany. Before being sold, the property has to be valuated accurately in advance. Unfortunately, this involves a number of manual tasks that need to be transparent and explainable, obeying laws imposed by the Federal Financial Supervisory Authority (BaFin).
In cooperation with the Deutscher Sparkassenverlag (DSV) group, we seek to understand the variation between the bidding and selling prices depending on various attributes of the property, such as its location or number of bath rooms. Further, we aim to analyze how well the valuation quality improves if we drop the restriction of the process being fully explainable with respect to many complex rules.
Given millions of real-estate valuation data and scientific freedom, we aim to enhance the current valuation process. As the core goal of the project, we want to investigate and understand the underlying structure of the data and learn relations not revealed by conventional methods. We want to compare the accuracy of these new, possibly unexplainable approaches to the valuations approved by the BaFin. A visual representation of this vision is depicted in the figure below.
Additionally, building on these insights, we want to create a user interface that allows for more fine-grained searches and dynamic changes on the fly. This shall result in a Germany-wide map portraying the real-estate prices given a certain configuration.