A Method for Predicting Termite Damage in Wooden Cultural Properties Using a Machine Learning Model

Abstract

In this study, a machine learning model was established using termite damage and meteorological data on wooden cultural properties, and the prediction performance was evaluated. The data were divided into termite damage data, the location of wooden cultural properties, and meteorological data. Three observatories were searched based on the location of the wooden cultural properties, and meteorological data for about 8 years from 2010 to 2018 were combined to make a total of 491 data sets. As a result, it was confirmed that the value of small evaporation as a meteorological factor that directly affects termite damage is the time series independent variable that best explains the model, and showed an accuracy of 72.8% when the Linear SVM algorithm model was used. The value of small evaporation is the synoptic meteorological data of the Korea Meteorological Administration, and it is collected only at specific stations, not meteorological factors observed at all stations. Therefore, It is difficult to obtain enough data to make a predictive model. Since machine learning models can improve accuracy when the number of data is sufficient, prediction performance can be improved if more termite damage data and the value of small evaporation are obtained.