Development of Typhoon Damage Prediction Function using Tukey's Ladder of Power Transformation

Abstract

In this study, we use multiple linear regression to develop a function for predicting damage caused by typhoons. The number of explanatory independent variables constituting the damage prediction function is large and varied because typhoon damage is attributed to a combination of factors such as heavy rain, strong wind, and waves. However, existing data of typhoon damage are insufficient for developing a damage prediction function. To resolve this problem, a model for prediction power only was developed, and the leave-one-out-cross-validation (LOOCV)-forward selection method was applied to select the variables. In addition, the power transformation of Tukey's Ladder of Powers was adopted to create linearity in the dependent and independent variables. The typhoon damage prediction function was developed for 22 districts with more than 16 datasets. The transformation factor λ, representing these 22 districts, was shown to vary according to the regional characteristics of each city and district. The best results occurred when λ = 0.2 and λ = 0.3 for the normalized root-mean-square error (NRMSE) and coefficient of determination (R2) standards, respectively. Keywords: Tukey’s Ladder of Power, Typhoon Damage, Prediction Function