An analysis of the relative variable importance to flood fatality using a machine learning approach

Abstract

Vietnam is regularly and severely affected by flood events and there were nearly 14,000 dead people in 200 separate floods from 1989 to 2015. However, there have been limited studies specifically on flood-related mortality in Vietnam. This paper presents a longitudinal investigation of flood fatalities in Vietnam. More specifically, we use the available national disaster database and machine learning techniques to investigate theimportance of different attributes of flood damage to the attribute of flood fatalities. The results show that thehousing damage attribute significantly influences the fatality attribute, of which the weights are 0.45, 0.62, and 0.36 for the random forest, boosting, and multiple linear regression techniques, respectively. Thus, it is recommended that the proper policy prioritize housing improvements, establish evacuation plans, and developa strategy for temporary flood shelters in flood-prone areas. Understanding how various components of flooddamage are more likely to lead to fatalities analyzed in this study is critical for developing risk reductionstrategies.