Feature-Weighting-Based Prediction of Drought Occurrence via Two-Stage Particle Swarm Optimization

Drought directly affects environmental sustainability. Predicting the drought at earliest opportunity will help to execute mitigation plans. Several indices are used predict severity of across different geographical regions. The two main in India for meteorological standardized precipitation index (SPI) and evapotranspiration (SPEI). This work is a study find ability above mentioned state Tamil Nadu using 62 years data. prediction results evaluated performance metrics precision, recall, f1 score, Matthews correlation coefficient, accuracy. dataset severely imbalanced due low number incidence years. Hence there exists tug war between precision so improving it without affecting one another, multi-objective optimization process applied. improved by filter-global-supervised feature weighting wrapper-global-supervised techniques. In filter-based approach, information gain measure Pearson coefficient as weights. For wrapper-based two-stage particle swarm (PSO) designed calculate weights features, random forest classifier. PSO constructs best population set individual objectives then searches around that multiple contradicting converge into solution easier. When compared classification weighting, achieves 45% improvement precision. However, only moderate recall obtained. According findings, SPI3 SPEI12 should be given more weightage metrological prediction.