Revealing the economic consequences of tornadoes using a two-stage SVM-based predictive approach

Abstract Limited studies have concentrated on estimating economic losses caused by tornadoes through combining socioeconomic and demographic vulnerability with geographic and terrain attributes. The current study built a two-stage support vector machine (SVM) based predictive tool for both determining the tornado-prone regions and the resultant economic losses. During the identification of tornado-prone regions, the SVM model achieved an accuracy of 80.57% and an area under the receiver operating characteristic (AUC) score of 86.79%, demonstrating its ability to classify the regions susceptible to experience tornadoes. Also, estimation of economic losses for the respective regions provided root mean squared error (RMSE), mean absolute error (MAE) and coefficient of determination (R 2 ) values of 0.5352, 0.4814, and 0.9175, respectively. The analysis showed that 17.59% of Texas is under very high economic loss risk. The game theoretical Shapley Additive explanation (SHAP) analysis further highlighted the contribution of Fujita scale of tornadoes, longitude, number of houses, and tornado area into the likelihood of tornado incidents. Hence, the holistic predictive framework introduced is useful in creating instruments to help government disaster management officials, professionals, and insurance companies, given the anticipated rise in the severity, frequency, and duration of extreme weather events in the context of global warming.