Incremental learning for property price estimation using location-based services and open data

Francisco Álvarez, Edgar Roman-Rangel, Luis V. Montiel

Abstract: This paper proposes a tree-based incremental-learning model to estimate house pricing using publicly available information on geography, city characteristics, transportation, and real estate for sale. Previous machine-learning models capture the marginal effects of property characteristics and location on prices using big datasets for training. In contrast, our scenario is constrained to small batches of data that become available in a daily basis, therefore our model learns from daily city data, employing incremental-learning to provide accurate price estimations each day. Our results show that property prices are highly influenced by the city characteristics and its connectivity, and that incremental models efficiently adapt to the nature of the house pricing estimation task.