Uso de la inteligencia artificial en la estimación del precio de la vivienda urbana en municipios intermedios de Colombia

Abstract

The real estate market in Colombia faces significant challenges owing to the lack of detailed and updated information, which makes it difficult to accurately estimate property prices. This limitation affects the ability of buyers and sellers to make informed decisions in real estate transactions, which in turn has a negative impact on the citizenry’s economic stability. In addition, factors such as unemployment and economic recession negatively affect people's financial perceptions and demand for housing. To address these issues, an artificial intelligence model combining a convolutional neural network and a direct feedforward was developed to estimate real estate prices in intermediate municipalities in Colombia. This model was based on the construction of an information extraction system using web scraping to collect data from real estate offers, formulation of a neural network model that integrates images and alphanumeric features, and evaluation of the model's performance using regularization techniques. During the development of the project, we identified the need to address the technical complexity of the web scraping process and optimize the hyperparameters of the neural network model to improve its generalization capability. We observed that certain hyperparameter configurations, such as intermediate learning rates and moderate regularization values, generated satisfactory results in the training and testing metrics. However, we also noted an inability to generalize new information when evaluating models with new datasets, suggesting the need to expand the size of the datasets and consider other artificial intelligence approaches