Predicting Damage Characteristics of Heavy Rain Disasters Using Artificial Neural Network

Abstract

Heavy rain disaster is a representative natural disaster in Korea, and causes considerable damage every year as a result of typhoons and heavy rain. Various preventive measures and reduction facilities have been introduced to reduce heavy rain damage, but damage continues to occur. Accurate damage prediction is necessary for preemptive response to heavy rain disasters. A prediction study of the damage characteristics of such disasters using machine learning was conducted. Artificial neural network (ANN) was applied as a machine learning technique, and damage data of the aforementioned disasters from 1999 to 2019 were used. Damage prediction using an ANN was analyzed in terms of total rainfall, maximum daily rainfall, and total damage amount. The prediction of the damage characteristics was most accurate for the total damage amount. Total and daily maximum rainfall are less correlated with the characteristics of various damage amounts, which may explain the low accuracy of the analysis.