Uso de machine learning para previsão de valores de apartamentos no município do Recife

Abstract

The COVID-19 pandemic has brought with it a series of economic effects and transformations related to behavior and the way people live, which, in turn, have had repercussions on property prices and real estate demand. In this context, property price forecasting assumes an extremely important role, contributing to more informed decisions, mitigating risks, and promoting greater transparency in the real estate sector. The implementation of automation in price forecasting further enhances this dynamic, significantly improving accuracy, efficiency, and reliability of predictions, while providing adaptability to economic fluctuations with greater agility. Utilizing listings available on OLX, a georeferenced database was created to generate a residential apartment price prediction model in Recife, using machine learning models in AutoML. This tool automates the development of machine learning models, enabling rapid experimentation and a focus on problem-solving. The work indicates that the poor geographical distribution of the data has biased the results of the models. Furthermore, it was concluded that the data found on online buying and selling platforms are insufficient for generating a machine learning model that achieves an acceptable level of accuracy in Recife, mainly because transaction values for the properties are not provided, only the advertised prices. However, this current work provides significant contributions to the advancement of research related to automation in real estate price prediction.