[3S3-IS-2e] Machine learning

[3S3-IS-2e-02] Incremental informational value of floorplans for rent price prediction

Applications of modern computer vision techniques in real-estate

〇Jiyan Jonas Schneider^1,2, Hoshino Takahiro^1,2 (1. Graduate School of Economics, Keio University, 2. AIP Center, RIKEN)

Regular

Keywords:Computer Vision, Real estate , Rent prices, Deep Learning, Hedonic Price model

This report examines whether a consideration of floorplan images of real-estate
apartments can effectively improve real-estate rental price predictions. We use
a modern computer vision technique to predict the rental price of apartments
using the floorplan of the apartment exclusively. Afterward, we use these
predictions combined with a more traditional hedonic pricing method to see
whether its predictions improved. We found that by including the predictions, we
were able to increase the accuracy of the predictions from an R² of
0.915 to an R² of 0.945. This improvement suggests that floorplans
contain considerable information about rent prices, not captured in the other
explanatory variables used. Further investigation, including more explanatory
variables about the apartment itself, could be used in future research to
further examine the price structure of real estate and better understand
consumer behavior.