Damásio, Bruno Miguel PintoNeves, DavidRodrigues, Filipe Alexandre Cleto dos Santos de Sousa2025-11-202025-11-202025-11-10http://hdl.handle.net/10362/191124Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Business AnalyticsEstimating real estate market value persists to be a complex problematic in public policy, investment, and taxation, particularly in countries like Portugal, where housing is central to both household wealth and government revenue. Traditional valuation models often rely on limited structural or locational data, reducing their accuracy. This thesis introduces a new approach using over a decade of administrative property and transaction tax records from the Portuguese National Institute of Statistics to predict housing prices through machine learning. A range of models, including random forest, gradient boosting, and traditional linear regression, were trained on a harmonized dataset, with extensive preprocessing to handle data complexity. Results show that non-linear ensemble models consistently outperform linear approaches, better capturing the diverse factors influencing price formation. Key variables included tax classification, official valuations, and location and time attributes. Model transparency was enhanced through interpretability techniques that revealed consistent, meaningful patterns. While not intended for precise individual price prediction, the models offer valuable insights for macroeconomic monitoring, fiscal planning, and evidence-based housing policy in Portugal.engReal EstateMachine LearningAdministrative Tax DataHousing MarketPredictive ModelingSDG 4 - Quality educationSDG 8 - Decent work and economic growthSDG 9 - Industry, innovation and infrastructureSDG 11 - Sustainable cities and communitiesmaster thesis204070651