Albuquerque, Carina Isabel AndradeHenriques, Roberto André PereiraLavinha, Mariana Ferraz2025-11-182025-11-04http://hdl.handle.net/10362/190948Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Business AnalyticsMachine learning (ML) has been increasingly recognized as a powerful tool in macroeconomic forecasting, with the potential to outperform traditional econometric methods. Accurate GDP forecasting is crucial for central banks, due to its significant influence on economic policy and financial stability. This thesis aims to evaluate the predictive performance of ML algorithms, namely Random Forest (RF), Support Vector Machines (SVM), and Long Short-Term Memory Neural Networks (LSTM), compared to the traditional ARIMA model for forecasting Portugal's GDP. The study uses economic time series data from Banco de Portugal incorporating eleven economic series covering the period from 2011 to 2023. The models were applied to both original and differenced time series, with differencing significantly improving performance across all algorithms. The results showed that ML models generally outperformed ARIMA model, with RF and SVM performing the best out of all tested models. RF was the most consistent model in multivariate forecasts achieving the most accurate prediction, while SVM performed better in the univariate setting, balancing robustness and forecasting precision. LSTM, on the other hand, underperformed, likely due to dataset limitations, suggesting that deep learning approaches may require larger volumes of data and greater model complexity to reach their full potential. The inclusion of additional explanatory variables notably improved the ML models’ performance, particularly for RF. The thesis demonstrates the potential of ML in macroeconomic forecasting and points to future research and exploration opportunities in this ongoing field.engGDPPortugalForecastingTime SeriesMachine LearningSDG 8 - Decent work and economic growthmaster thesis204070864