Comparison of Machine Learning Models for Drought Prediction in Central Myanmar

Drought is a highly dangerous natural hazard that causes water shortages, crop production losses, and environmental fires. Droughts are distinguished by meteorological metrics such as the standardized precipitation index (SPI). SPI is the most general and simple way to compute its values using only precipitation data. A popular way to determine the SPI is to fit the gamma function to rainfall data. The SPI of 14 sites was computed using monthly rainfall data over 33 years (1990--2022). The current study analyzed how well two well-known machine learning models, Random Forest (RF) and Decision Tree (DT), predicted the standardized precipitation index (SPI) across several time periods such as (1, 3, 6, 9, 12, 24) months. The models were built at the Nyaung Oo station in Myanmar’s drought-prone central region. Results proved that the proposed models could effectively predict droughts at the Nyaung Oo station. The Decision Tree model and Random Forest model recognized the SPI-12 values of Nyaung Oo station. In 80% of training and 20% of testing datasets, the RF model predicted the SPI better than the DT model, with a Root Mean Square Error (RMSE) of 0.19, a Mean Square Error (MSE) of 0.04, and the highest coefficient of determination ($R^{\mathbf{2}}$) of 0.97. The measurement accuracy is 0.96.