Machine Learning and Data-Driven Approaches in Spatial Statistics: A Case Study of Housing Price Estimation

Abstract

The intertwining of socio-spatial complexity with that of price formation leads to highly challenging questions when modeling real estate markets and the dynamics of property prices. The exact same apartment typically will not have the same price depending on its location in the city - due to specifics of the neighborhoods and even micro-neighborhoods that are difficult to quantify. Traditional methods rely on the so-called hedonic approaches modified to incorporate spatial effects via geographically weighted regressions. However, the recent availability of big data pertaining to the socio-economic characteristics of cities, at a very fine-grained level, should allow one to capture in much finer detail the complex relationship between space and price in the real estate market. Our approach is two-fold, we first apply a simple Self-Organizing Map (Kohonen) algorithm on vast sets of demographical, economical and infrastructural data in order to bring out the socio-spatial structure of a city and then use this cluster information into the spatial diffusion process of the GWR. SOM gives a notion of proximity between clusters and thus provides a multi-scale degree of similarity (unlike other clustering algorithms like k-mean).