Real State Price Estimation in Brazil Using Machine Learning

The primary goal of this research is to develop a predictive model for real estate prices specifically tailored to Florianopolis, SC. The process involved extracting data from a real estate platform using web scraping techniques. Following this, the collected data underwent extensive cleaning, organization, preprocessing, and classification before being employed in three predictive machine learning models: Random Forest, Lasso Regression, and XGBoost. The models were trained using various attributes including the number of bedrooms, bathrooms, parking spaces, total area of the property (in square meters), neighborhood, property type, and the presence of a condominium. To enhance accuracy and minimize errors, cross-validation and hyperparameter optimization techniques were applied. Among the models tested, XGBoost was the top performer, showcasing an RMSE (Root Mean Square Error) of 0.16 and an R2 of 0.98. The model's accuracy and predictive capabilities hold significant potential for real estate agents, investors, and individuals seeking insights into property pricing trends in Florianopolis and generalizable to Brazil. The findings from this study highlight the effectiveness of employing machine learning techniques in predicting real estate prices. Moreover, the model's success underscores its practical applications in guiding investment decisions, aiding in property valuation, and providing valuable insights into the dynamic real estate market of Florianopolis, SC.