Abstract:
Urban scholars have made great advances to understand the reciprocal relations between households and their immediate environments as a means for the creation of efficient urban administrative systems. However, from an urban management perspective, reliance on geographical areas fixed for long periods of time as basic data collection constitutes a problem. Modern urban areas are in a permanent state of flux because of changing preferences, willingness to pay, location choices, and physical development. In this constantly changing context, what is the most appropriate delimitation of a "neighborhood", defined as a small and relatively homogeneous area in a certain (and temporary) urban configuration? This paper contributes to the growing literature on the use of data analytic tools in urban studies and neighborhood delimitation in housing sub-markets, exploiting big data on real-estate transactions in England and Wales during a long period of time. The results shed light on the importance of organic urban features and the drawbacks of rigid geometric definitions. They also highlight the importance of the usage of deep Machine Learning (ML) tools such as Artificial Neural Network (ANN), alongside with traditional methods. The paper's contribution to urban governance is the suggestion of a smart and dynamic system aimed at defining the most appropriate areas for urban management given a specific period and situation. The suggested framework can be implemented periodically, helping to define homogeneous spatial units (neighborhoods) with large variances among them, allowing for designing urban policies tailored to each one of them.