Real Estate Price Prediction Using Spatial Artificial Neural Networks: An Alternative to Market Value for Taxation of Urban Real Estate

• DOI: https://doi.org/10.22533/at.ed.2164172419082

• Palavras-chave: GWANN, ANN, Accuracy, Spatial Data, Non-Spatial Data.

• Keywords: GWANN, ANN, Accuracy, Spatial Data, Non-Spatial Data.

• Abstract:

  The spatial issue is important for many empirical problems. The location of properties in relation to their price (or market value) for property taxation is an example and requires efficient methodologies for its prediction. This study was dedicated to comparing two neural network methodologies for predicting the market value of properties that incorporate the location of the properties in different ways. To this end, a sample of property offers for the city of Curitiba/Brazil was used. In one model, using GWANN, the location was attributed through the latitude and longitude of the properties, intrinsic characteristics of the property and other locational characteristics such as distances to points of interest (POIs). The other model (RNAD), a traditional ANN, differed from the first by considering the neighborhood where the properties are located.