Development of the Model for Disaster Occurrence Prediction Using a Model Based Machine Learning
By Dong-in Park, et al.
The frequency of natural disasters such as typhoons and heavy rains is increasing, and the damage to disasters is also increasing. The Korea Meteorological Administration, which conducts weather and climate observation, studies, and forecasts, issues that do not consider regional characteristics by uniformly issuing spcial weather reports standards nationwide.
Therefore, this study develops a model to establish regional heavy rain damage standards and determine whether heavy rain damage occurs by linking rainfall data and heavy rain damage data in the metropolitan area (Seoul Metropolitan Government, Incheon Metropolitan City, Gyeonggi Province). In this study, the probability that probability variables are included in the probability density function and the section for the rainfall and heavy rain damage data was calculated. Based on this, a cumulative distribution function (CDF) was used to calculate the probability that a given probability variable is less than or equal to a specific value. Based on this, rainfall standards were established according to the four stages of disaster crisis warning (blue, yellow, orange, and red) presented in the Framework Act on Disaster and Safety Management by judging the occurrence of heavy rain damage. To develop an optimal decision tree (DT), support vector machine (SVM), random forest (RF), and XGBoost (extreme gradient boost) model based on the set criteria, a total of four models were developed. As a result of accuracy evaluation using F1 Score for each model, the F1-Score of the random forest model was the highest at 0.49. The XGBoost model showed the best accuracy among the applied models. Based on the machine learning-based heavy rain damage prediction model presented in this study, risk information on heavy rain damage can be provided, which can be used as basic data for disaster managers' decision-making.
Key words : Machine Learning, AI-Based Model, Special Weather Reports
Development of the Model for Disaster Occurrence Prediction Using a Model Based Machine Learning
By Dong-in Park, et al.
The frequency of natural disasters such as typhoons and heavy rains is increasing, and the damage to disasters is also increasing. The Korea Meteorological Administration, which conducts weather and climate observation, studies, and forecasts, issues that do not consider regional characteristics by uniformly issuing spcial weather reports standards nationwide.
Therefore, this study develops a model to establish regional heavy rain damage standards and determine whether heavy rain damage occurs by linking rainfall data and heavy rain damage data in the metropolitan area (Seoul Metropolitan Government, Incheon Metropolitan City, Gyeonggi Province). In this study, the probability that probability variables are included in the probability density function and the section for the rainfall and heavy rain damage data was calculated. Based on this, a cumulative distribution function (CDF) was used to calculate the probability that a given probability variable is less than or equal to a specific value. Based on this, rainfall standards were established according to the four stages of disaster crisis warning (blue, yellow, orange, and red) presented in the Framework Act on Disaster and Safety Management by judging the occurrence of heavy rain damage. To develop an optimal decision tree (DT), support vector machine (SVM), random forest (RF), and XGBoost (extreme gradient boost) model based on the set criteria, a total of four models were developed. As a result of accuracy evaluation using F1 Score for each model, the F1-Score of the random forest model was the highest at 0.49. The XGBoost model showed the best accuracy among the applied models. Based on the machine learning-based heavy rain damage prediction model presented in this study, risk information on heavy rain damage can be provided, which can be used as basic data for disaster managers' decision-making.
Key words : Machine Learning, AI-Based Model, Special Weather Reports