Predicting Drought Based on Weather and Soil Data Analysis

Drought is one of the major natural disasters that have a long-term impact. Its devastating effects can be lessened by forecasting drought properly and implementing suitable prevention and disaster reduction strategies. Though prior research has solved this problem to some extent, the forecasting drought still lacks precision. This work provides a comprehensive drought prediction technique based on many machine learning models. This paper considers a variety of significant characteristics, including historical climatic data, wind speed, rainfall, and soil moisture. Because the data was uneven, a Synthetic Minority Oversampling Technique (SMOTE) was used to up sampling of the models. Models that were first up sampled with SMOTE analysis achieved an average accuracy of 0.78. The KNN model with SMOTE analysis has the highest accuracy, 0.7955, and an F1-score of 0.798.