Decoding Drought: Embracing Simplicity in Effective Predictive Models

Drought, a recurring environmental challenge, poses severe threats to ecosystems, agriculture, and water resources worldwide, significantly affecting the ecological balance and socioeconomic stability. The state of Rajasthan, with its long history of droughts, faces these issues due to its arid climate and irregular rainfall. This paper presents a comparative analysis of drought prediction models, focusing on the Rajasthan region using two datasets: Vegetation Health Index (VHI) and VHI + Rainfall. The study evaluates six machine learning and deep learning models, including Multivariate Linear Regression (MLR), Random Forest (RF), Multilayer Perceptron (MLP), K-Nearest Neighbors (KNN), Support Vector Regressor (SVR), and 1-Dimensional Convolutional Neural Network (1DCNN). The integration of two datasets contributes to the study's novelty, with KNN excelling in incorporating both VHI and Rainfall. MLR emerges as a robust choice with the highest average R2 score of 0.944, and lowest average MSE and MAE of 0.004 and 0.033 respectively for the VHI dataset while KNN leads with an average R2 score of 0.928, and least MSE and MAE of 0.005 and 0.036 respectively for VHI + Rainfall dataset. These findings emphasize the potential of simpler models for effective drought prediction.