A Machine Learning Framework for Drought Prediction in Tropical Malaysia With Input Data Uncertainty

ABSTRACT The far‐reaching and devastating impacts of drought underscored the need for effective drought forecasting. This study addressed the critical gap of the often‐overlooked impacts of input data uncertainty on the forecasting accuracy. The study incorporated the input data uncertainty as a novel approach into the hybrid Extreme Gradient Boosting‐Recurrent Neural Network (XGBoost‐RNN) to forecast Standardised Precipitation Index (SPI) and Standardised Precipitation Evapotranspiration Index (SPEI) across different time scales in Malaysia. The results were assessed using coefficient of correlation ( R ), root mean square error (RMSE), mean absolute error (MAE), mean square error (MSE), Nash Sutcliffe Efficiency (NSE) and Wilmott's Index (WI). The XGBoost‐RNN incorporating input data uncertainty (XGBoost‐RNN‐ID) outperformed its counterpart, achieving R values up to 0.9988, WI up to 0.9994, and RSME, MAE, and MSE as low as 0.0466, 0.0367, and 0.0022. The NSE values categorised XGBoost‐RNN‐ID as “very good” across all time scales, demonstrating its reliability in drought forecasting.