<![CDATA[This study aimed to develop and evaluate a hybrid forecasting model integrating Ant Colony Optimization (ACO), Backpropagation Neural Network (BPNN), and XGBoost for monthly rainfall prediction. The proposed hybrid framework combines optimization, neural network, and boosting techniques within a single forecasting model. Monthly rainfall time-series data from 2016 to 2025 in Alas Subdistrict, Sumbawa Regency, West Nusa Tenggara, were used, comprising 120 observations obtained from BPS, BMKG, and NASA POWER. The methodology included data preprocessing, an 80%–20% chronological training–testing split, model development, and performance evaluation using MSE, MAE, RMSE, MAPE, and R². The results indicated that all models experienced performance degradation during testing, suggesting overfitting and limited generalization capability. The testing RMSE values for ACO, BPNN, XGBoost, and the hybrid model were 108.41, 109.33, 109.21, and 106.08 mm, respectively. The corresponding testing MAPE values were 2836.5%, 2900.5%, 2073.9%, and 2602.2%, although these values should be interpreted cautiously because rainfall observations occasionally approached zero. While the hybrid model achieved the lowest testing RMSE, the improvement over the best individual model was modest, and all testing R² values remained negative, indicating weak generalization capability. Therefore, the findings should be regarded as preliminary, and further validation using larger datasets, exogenous climatic predictors, baseline forecasting methods, and more rigorous evaluation procedures is required.]]>
