Browsing by Author "Karanko, Lauri"

Determining the optimal rental price of an apartment is typically something that requires a real estate agent to gauge the external and internal features of the apartment, and similar apartments in the vicinity of the one being examined. Hedonic pricing models that rely on regression are commonplace, but those that employ state of the art machine learning methods are still not widespread. The purpose of this thesis is to investigate an optimal machine learning method for predicting property rent prices for apartments in the Greater Helsinki area. The project was carried out at the behest of a client in the real estate investing business. We review what external and inherent apartment features are the most suitable for making predictions, and engineer additional features that result in predictions with the least error within the Greater Helsinki area. Combining public demographic data from Tilastokeskus (Statistics Finland) and data from the online broker Oikotie Oy gives rise to a model that is comparable to contemporary commercial solutions offered in Finland. Using inverse distance weighting to interpolate and generate a price for the coordinates of the new apartment was also found to be crucial in developing an performant model. After reviewing models, the gradient boosting algorithm XGBoost was noted to fare the best for this regression task.