A Hybrid Model of Machine Learning with Web Scrapping to Determine the Urban Dynamics of the Constructions of Medellín - Colombia

Abstract

The cities growth has its main reason that more than 50% of the world's population lives in urban areas and in Latin America that percentage exceeds 75%. In many aspects, this growth is studied as a territory-oriented manner, increasing the area of occupation only in two dimensions; cities grow in territory and volume. The present research focuses on vertical growth from a view composed of two independent models, the first uses a classification Machine Learning model using statistical values of the last ten years of changes in story levels and other variables to predict if a property would present a vertical growth, the other model is created with information taken from the extraction using Web Scrapping technics from the market offers from a popular real state web page in Colombia. These two data are projected geographically using the neighborhoods of Medellín and through a raster model of average density, the changes are added to provide a result by neighborhoods of the city of Medellín where there will be the highest probability of new stories or new buildings, the goal is to improve the management of the territory in terms of contributing to the study of the city and the fiscal and cadastral aspects of the territory.