Management in drought-prone areas is an accurate predictor of agricultural drought. It involves carefully monitoring and predicting drought conditions, which negatively affects plant growth and crop yield. Much research effort has been devoted to developing systems and models for predicting drought in agriculture. Droughts are natural weather patterns that can last for seasons. The severity can range from short-term to long-term. Farming households face challenges in meeting their needs due to high agricultural costs, hindering national development. This study uses region-specific climate and meaningful geographic data to predict drought length and severity over time. Drought severity is measured on a scale from 0 to 5, with 0 representing the most severe condition and 5 indicating the most severe. A variety of factors contributing to drought breeding are identified at the beginning. These features are used to train multivariate time series models such as Prophet, VAR (vector auto-regression), LSTM (long-term short-term memory), and comparison of true and predicted values ​​The results were encouraging. In a pilot study comparing several machine learning algorithms for agricultural drought forecasting, the LSTM model outperformed the VAR and prophet models.