Comparative Assessment of Improved SVM Method under Different Kernel Functions for Predicting Multi-scale Drought Index

This paper focus on the drought monitoring and forecasting for semi-arid region based various machine learning models SPI index. Drought phenomena are crucial role in agriculture drinking purposes area. In this study, Standardized Precipitation Index (SPI) was used to predicted future upper Godavari River basin, India. We have selected ten input combinations of ML model were prediction three timescales (i.e., -3, SPI-6, SPI-12). The historical data from 2000 2019 creation prediction, these datasets divided into training (75% data) testing (25% models. best subset regression method sensitivity analysis applied estimate most effective variables estimation 3, 6, 12. improved support vector using sequential minimal optimization (SVM-SMO) with kernel functions i.e., SMO-SVM poly kernel, Normalized PUK (Pearson Universal Kernel) RBF (radial basis function) developed SPI-3,6 12 months. accuracy compared statistical indicators root mean square error (RMSE), absolute (MAE), relative (RAE), squared (RRSE), correlation coefficient (r). results study area been showed that precisely SPI-3 (R2 = 0.819) SPI-12 0.968) values at Paithan station; 0.736) SPI-6 0.841) Silload station, respectively. is found 0.846) station 0.975) station. SVM-SMO observed, (i.e. SPI-12), while good both stations. ability algorithm successfully multiscale under climate changes. It can be helpful decision making water resource management tackle droughts central