Rainfall forecasting using extreme gradient boosting

Abstract

Rainfall is an important factor affecting the agricultural sector, especially in areas such as Grobogan Regency, Central Java, which is one of the national rice barns. Rainfall uncertainty often leads to crop failure due to drought or flooding, which is caused by inaccurate rainfall predictions. This research aims to build a dasarian rainfall forecasting model using the Extreme Gradient Boosting (XGBoost) method. The data used is ten-day (decadal) rainfall data from January 1991 to February 2025. The research steps start from data preprocessing which includes lag and seasonal feature creation, hyperparameter determination and hyperparameter tuning, and model evaluation. After conducting hyperparameter tuning to find the best model, the best model was found with n_estimator 448, learning rate 0.14, max_depth 3, min_child_weight 1, subsample 0.71, colsample_bytree 0.97, gamma 5.1, alpha 3.8, and lambda 8.89. The model formed is then evaluated using SMAPE (Symmetrical Mean Absolute Error) for test data of 36.48%. From the model formed, forecasting is carried out and the forecasting results can be utilized as a basis for decision making in agricultural planning.