Accurate streamflow foecasting in O'ahu Island, Hawaii, presents significant challenges owing to the island's volcanic terrain and extreme microclimates, ranging from arid to humid zones. In this study, we propose a novel solution to forecast discharge in four streams across O'ahu Island, namely, the Waikele, Waimea, Makua, and Kahana streams, which are located in arid, semiarid, subhumid, and humid climate conditions, respectively. Our innovative approach employs the honey badger algorithm (HBA) to optimize the multilayer perceptron (MLP) model. By classifying O'ahu's climate zones using the aridity index, the HBA-MLP model was used to forecast streamflow across diverse climatic conditions, a task that had not been previously undertaken because of the complexity of the hydrological system. The model demonstrated exceptional performance in forecasting the streamflow time series of the four climate zones. In the Waikele stream (arid zone) the model achieved a root mean square error (RMSE) of 0.80 m3/s, Kling-Gupta efficiency (KGE) of 0.95, and Nash-Sutcliffe efficiency (NSE) of 0.92. Similarly, in the Waimea stream (semi-arid zone), the model achieved an RMSE of 0.60 m3/s and coefficient of determination (R2) of 0.93, indicating accurate forecasting. In the Kahana stream (humid zone), the model achieved moderate success, with an RMSE of 0.22 m3/s and KGE of 0.85. However, the Makua stream (subhumid zone) presented more challenges, where the model attained an RMSE of 0.08 and KGE of 0.68, indicating that there are areas for improvement in streamflow forecasting. In this zone, Ultisols, which are characterized by shallow clay-enriched subsoils and a tendency for rapid subsurface flow, contribute to substantial variability in streamflow, which complicates accurate forecasting. In terms of the peak flow values, the model attained a mean absolute relative error (ARE) of 17.64 and 14.44 % for the arid and semiarid zones, respectively, indicating reliable forecasting of extreme flow events. The model achieved moderate accuracy for the humid zone with a mean ARE of 28.78 %. However, the model experienced greater difficulties for the subhumid zone, with a higher mean ARE of 29.38 %, reflecting the complexity of accurately predicting peak flows in highly variable climates. This study demonstrates the effectiveness of advanced machine learning models, such as the HBA-MLP model, in solving complex hydrological problems and offers valuable insights for improving real-time flood forecasting and water resource management in Hawaii. As the first comprehensive study to forecast streamflow across O'ahu's climate zones, this work fills a critical gap in regional forecasting research and highlights areas for future improvement, particularly in subhumid regions where complex subsurface flow dynamics and soil characteristics affects forecasting performance.