This is a project prepared during the Digital Economy and Decision Analytics - Blockchain and Cryptocurrency Seminar class at Humboldt University of Berlin. Check out the presentation in this repo
Some of the graphs form the research:
All polygons used:
All multipolygons used:
Distribution between areas of the following infrastructural objects:
Gyms, sports objects, etc.
Restaurants, bars, etc.
Sightseeing, museums, etc.
Cinemas, night clubs, etc.
Doctors, pharmacies, etc.
Kitas, playgrounds, etc.
Parks, dog parks, etc.
Shops, grocery stores, etc.
S-Bahn, U-Bahn, etc.
Initial model used for apartment prices:
Cleaned model:
Diagnostic plots:
Distance measurement, scoring and normalization. The distance measurement chosen for this project is the Euclidian distance. The following formula was used to calculate the scores:
Here 10 units (meters) are added to the distance in order to set the maximum score achievable for the measurement. After that the scores are rescaled via min-max normalisation in order to get the weights:
Here 1/2 is added in order to make the minimum weight 0.5 and the maximum 1.5 (half-price discount or extra payment)
The results:
Evaluating the landlord premiums:
Key takeaways:
- The size of the apartment, the presence of a kitchen, and a lift can be used as a solid foundation for a price check for the rent prices of apartments
- Infrastructural-wise Berlin tends to be a very single-centroid oriented city with few other areas of concentration emerging from different infrastructural objects
- Overall in Berlin, rent prices are not under- or overvalued based on the model used in this research, prices seem to be landlord premiums seem to be fairly normally distributed
- Overvalued properties seem to be more concentrated further outside the city centre and the more undervalued properties are located closer to the city center which can point to the lack of prestige living in the centre of the city