A disaster‐severity assessment DSS comparative analysis

This paper aims to provide a comparative analysis of fuzzy rule‐based systems and some standard statistical and other machine learning techniques in the context of the development of a decision support system (DSS) for the assessment of the severity of natural disasters. This DSS, which will be referred to as SEDD, has been proposed by the authors to help decision makers inside those Non‐Governmental Organizations (NGOs) concerned with the design and implementation of international operations of humanitarian response to disasters. SEDD enables a relatively highly accurate and interpretable assessment on the consequences of almost every potential disaster scenario to be obtained through a set of easily accessible information about that disaster scenario and historical data about similar ones. Thus, although SEDD’s methodology is rather sophisticated, its data requirements are small, which, therefore, enables its use in the context of NGOs and countries requiring humanitarian aid. In this sense, SEDD opposes to some current tools which focuses on one phenomena‐one place disaster scenarios (earthquakes in California, hurricanes in Florida, etc.) and/or have extensive and/or technologically sophisticated data requirements (real‐time remote sensing information, exhaustive building census, etc.). Moreover, although focused on disaster response, SEDD can also be useful in other phases of disaster management, as disaster mitigation or preparedness. Particularly, the predictive accuracy and interpretability of SEDD fuzzy methodology is compared here in a disaster severity assessment context with those of multiple linear regression, linear discriminant analysis, classification trees and support vector machines. After an extensive validation over the EM‐DAT disaster database, it is concluded that SEDD outperforms the methods above in the task of simultaneously providing an accurate and interpretable inference tool for the evaluation of the consequences of disasters.