Lavado, SusanaPereira, GustavoKunnemann, Hendrik2025-08-042025-08-042025-01-292025-01-29http://hdl.handle.net/10362/185994This study extends a time series forecasting project (PBL) on a small dataset by exam ining ensemble learning, including homogeneous (bagging) and heterogeneous (Dynamic Integration) approaches. While bagging slightly reduces accuracy (MAPE), it improves stability. By incorporating a novel error-based dynamic pairwise correlation strategy to enhance diversity between base-learners, the Dynamic Weighting with Selection method within Dynamic Integration significantly outperforms the baseline, reducing the error met ric MAPE by nearly 10% and the stability metric by over 20%. These findings highlight the effectiveness of ensemble learning, particularly DWS, for accurate and reliable forecasting in small datasets.engTime series forecastingEnsemble methodsBaggingMoving block boot-strapDynamic integrationDiversity among base-learnersSmall datasetmaster thesis203962931