An Artificial Neural Network-based Model for Estimating Damage of Natural Disasters

Abstract

In the previous research, we developed a regression model for estimating damage of natural disasters based on the public database. Although this model considers nonlinearities among variables by using log transformation for the dependent variable, it reveals limitations in improving estimation accuracy because of its inherent characteristics of linearity assumption between independent and dependent variables. In this study, we proposed an artificial neural network (ANN) based model to predict the amount of damages due to natural disasters more accurately, which does not require the linearity assumption among the variables. For verification of the proposed model, we compared the model estimates with those from the regression model, including the Natural Disaster Risk Index (NDRI), Regional Safety Grades (RSG), and actual damage amounts. According to the results of analysis, we can confirm that the estimates from the ANN-based model reveal a higher correlation with the actual damage amounts than those from the regression model or the assessment results of NDRI and RSG.