Modelos de aprendizagem de máquina para precificação de imóveis na cidade de Fortaleza

Abstract

Real estate transactions are carried out with the participation of various agents. Far beyond buyers and sellers, funders and government agencies impact and are impacted by their result and effectiveness of those transactions. All agents always aim to minimize the risk they face in this mode of operation since it represents a sum of about tens or hundreds of thousands of reais. Long before credit studies, real estate pricing studies may or may not lead to the sale of the property. Evaluation methodologies present in technical standards mainly address inspection aspects or, in some cases, the statistical analysis of the property and its surroundings. With the advances in computer science and statistical theories, new evaluation methods have emerged, such as machine learning, which has shown good results as oracle of values, reaching, in some cases, average errors close to or less than 6%. Such low error rates become quite influential in the market since they bring security to all parties during the purchasing process. This paper proposes to discuss the insertion of several assisted computational learning methodologies to real estate valuation, discuss their results and choose which one presents the best results for the city of Fortaleza. The data for the survey were provided by the Municipal Finance Department of Fortaleza. They contain information on transaction values and the real estate physical data, which will be the starting point for each of the training and forecasting models addressed in the body of this paper.