The attractiveness of the property is one of the most interesting, yet challenging category to model. Image characteristics describe certain attributes, examining the influence of visual factors on a price or a timeframe of the listing. In this project, the proposed are a set of techniques for a visual features extraction as an efficient numeric inclusion in a modern-day predictive algorithms. After comparing techniques in respect to all property-related images (indoor, outdoor, and satellite), the authors conclude that all techniques are efficient single-digit visual measures for housing price predictions.
The set of chosen image features carries significant amount of predictive power and over-performs some of the strongest metadata predictors. Without any need to replace a human exert in a real-estate appraisal process, we conclude that techniques presented in this paper could efficiently describe visible characteristics, bringing perceived attractiveness as quantitative measure into a housing predictive modeling. This project is currently under review for publication.