Dengue remains a significant public health threat in tropical regions, especially in Ecuador. We develop and validate a predictive model to anticipate dengue outbreaks at the provincial level in the country, utilizing machine learning techniques applied to a comprehensive dataset that combines epidemiological, environmental, and demographic information from official sources, including the Ministry of Public Health, INEC, NOAA, and Google Earth Engine. A structured database was built to facilitate the training and evaluation of several algorithms commonly used to predict dengue outbreaks, including XGBoost, Random Forest, and SVM, utilizing cross-validation and metrics such as accuracy, F1-score, and AUC-ROC. The results showed that the XGBoost model was the most effective. Risk labels (Low/Moderate/High) were derived from weekly incidence using p70/p90 cutoffs computed only for the 2019–2023 training period, with a time-series cross-validation pipeline that included province one-hot encoding and train-only SMOTE. Then, we used data from the year 2024 to determine precision–recall thresholds, prioritizing sensitivity for the High-risk class. The model achieved an accuracy of 0.794 and a macro-F1 score of 0.756 on the 2025 hold-out. For the High-risk class, the precision, recall, and F1 score were 0.714, 0.989, and 0.829, respectively. In addition, an interactive simulator was designed along with risk maps created using the Python Folium library, and forecasts for 2025 were obtained using the Prophet library to facilitate spatial visualization and support public health decision-making. This model represents a valuable and scalable tool for strengthening epidemiological surveillance and preventive strategies in Ecuador, thereby contributing to mitigating the impact of future dengue outbreaks in the country.