A Recent Trends in Drought Prediction: A Comparative Study of Machine Learning and Deep Learning Models (2020–2025)

Drought has emerged as a critical environmental challenge, significantly affecting agriculture, water resources, and socio-economic stability. Between 2020 and 2025, utilizing deep learning (DL) and machine learning (ML) methods has expanded rapidly for drought prediction. Commonly applied models include Long Short-Term Memory (LSTM), XGBoost, Random Forest (RF), Support Vector Machines (SVM), CNN–LSTM hybrids, and emerging Transformer-based architectures. This review presents a comparative analysis of recent literature, focusing on datasets, drought indices (SPI, SPEI, NDVI), and performance metrics, including R², RMSE, and accuracy. Results indicate that RF and SVM remain effective for short-term drought forecasting, while LSTM and hybrid DL models show superior performance for long-term predictions. Looking ahead, integrating Transformer-based hybrid frameworks with satellite-derived indices offers a promising direction for more accurate and reliable drought monitoring systems.