Typhoon Damage Forecasting with Self-Organizing Maps, Multiple Regression and Decision Trees

Damage caused by typhoons to both people and structures has decreased in Japan due to improvements of countermeasures against natural disasters, however, such damage still occurs. A typhoon warning that represents the risk posed by a typhoon with high accuracy should be issued appropriately. Thus, we propose a new typhoon warning system which forecasts the likely extent of damage associated with a typhoon towards humans and buildings. The relation between typhoon data and damage data is investigated and typhoon damage is forecast using typhoon data. Self-organizing maps (SOM), multiple regression analysis and decision trees were used for typhoon damage forecasting. We consider two types of forecasting: two-class (yes or no) and three-class (small, medium or large scale) damage forecasting. Experimental results on accuracy of two-class and three-class forecasting with SOM were 93.3% and 96.8%, respectively. The accuracy with SOM was much better than that with multiple regression and decision trees. We recommend a new typhoon damage forecasting method based on these results.