A Comparative Analysis of Machine Learning Algorithms for Agricultural Drought Forecasting

Abstract

The occurrence of drought is a climatic feature and is a phenomenon that happens over time. Depending on the severity, it can last for a short or long time. Farming households are trying to meet due to rising agricultural operating costs that hinder the country's development. This study aims to forecast the severity of the drought over time. Drought scores vary from 0 to 5, with 0 and 5 indicating the least and highest intensity drought conditions. This is done using weather and soil data of a region consisting of Precipitation, Surface Pressure, Humidity, Temperature, Wind Speed, and Soil data. The main reasons for the cause of drought are first identified. These features are used to train the multivariate time series models like Prophet, VAR (Vector Auto-Regression), LSTM (Long short-term memory), and Comparison of actual v/s predicted values. The results were promising. The study has done an analysis comparing different machine learning algorithms for agricultural drought forecasting and it was found that the LSTM model performed better than VAR and Prophet models.