This study presents a reproducible ensemble machine learning approach for forecasting dengue outbreaks. The model, developed for Brazil, integrates spatial and temporal data to predict dengue incidence rates (DIR) one month ahead at the state level, with a focus on children under 19 years old. Comparative analyses with a dummy model and ablation studies demonstrate the model's effectiveness across 27 Brazilian Federal Units. The approach was also tested in Peru, showing its transferability to different epidemiological contexts. The model's success in Brazil and Peru highlights its scalability and practical application in public health. The study emphasizes the importance of integrating climate and socio-economic data to improve dengue forecasting, and it contributes to climate services for health by identifying factors triggering outbreaks. The model's innovation lies in its application to data-scarce environments and its ability to handle complex, multi-modal data. The ensemble approach combines CatBoost, SVM, and LSTM models, with results validated through extensive testing. The model's ability to predict DIR with high accuracy and low uncertainty, particularly in regions with stable seasonality, underscores its effectiveness. The study also highlights the importance of data preprocessing, including normalization and dimensionality reduction, to enhance model performance. The results demonstrate the model's ability to forecast dengue outbreaks accurately, providing valuable insights for public health decision-making. The study underscores the need for interdisciplinary collaboration in addressing global health challenges, emphasizing the role of advanced analytics in public health operational frameworks.This study presents a reproducible ensemble machine learning approach for forecasting dengue outbreaks. The model, developed for Brazil, integrates spatial and temporal data to predict dengue incidence rates (DIR) one month ahead at the state level, with a focus on children under 19 years old. Comparative analyses with a dummy model and ablation studies demonstrate the model's effectiveness across 27 Brazilian Federal Units. The approach was also tested in Peru, showing its transferability to different epidemiological contexts. The model's success in Brazil and Peru highlights its scalability and practical application in public health. The study emphasizes the importance of integrating climate and socio-economic data to improve dengue forecasting, and it contributes to climate services for health by identifying factors triggering outbreaks. The model's innovation lies in its application to data-scarce environments and its ability to handle complex, multi-modal data. The ensemble approach combines CatBoost, SVM, and LSTM models, with results validated through extensive testing. The model's ability to predict DIR with high accuracy and low uncertainty, particularly in regions with stable seasonality, underscores its effectiveness. The study also highlights the importance of data preprocessing, including normalization and dimensionality reduction, to enhance model performance. The results demonstrate the model's ability to forecast dengue outbreaks accurately, providing valuable insights for public health decision-making. The study underscores the need for interdisciplinary collaboration in addressing global health challenges, emphasizing the role of advanced analytics in public health operational frameworks.