This study evaluates the performance of ensemble machine learning models for predicting weekly dengue incidence in Zamboanga Sibugay using a 13-year dataset of DOH-reported dengue cases and NASA POWER climate variables. Five ensemble algorithms, namely Random Forest, XGBoost, LightGBM, CatBoost, and Decision Tree, were assessed against multiple error metrics including Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Squared Error (MSE). Among these, Random Forest yielded the highest predictive accuracy, achieving R2 scores of 0.827 (validation) and 0.715 (test), with corresponding MAE of 1.52, RMSE of 2.02 and MSE of 4.98. The models effectively captured complex, non-linear relationships between temporal indicators such as morbidity week and key climate features like precipitation and humidity among others. Spatial analysis revealed consistently high dengue case counts in municipalities such as Ipil, Siay, and Titay. Due to the inherently non-linear nature of dengue transmission dynamics, individual climate variables showed weak linear correlations with case counts; however, model performance was significantly enhanced when spatial and temporal dimensions were integrated. These findings highlight the potential of ensemble models as data-driven forecasting tools, offering a foundation for future integration into early warning systems and public health decision-support platforms in climate-sensitive regions.