A Machine Learning Approach to Analyze Natural Hazards Accidents Scenarios

Abstract

Climate change has contributed to an increasing frequency and severity of natural hazards accidents over recent years, and the increasing trend is expected to continue and escalate. Globally, demographics are changing and urbanization, population growth and increasing coastal populations make societies more exposed and vulnerable to extreme weather events. As a consequence, attention towards natural disasters is increasing along with the interest in approaches to manage emerging risks. Some industries have been experiencing major losses to hazards, while others might be hit harder in the future. Current research shows that there is a need to further investigate underlying reasons for variations in disaster timing, impacts, and outcomes, as well as mitigation strategies. The purpose of this research is to enhance the understanding of natural disaster mortality and unravel underlying causes and influential factors that can inform decision-making and be relevant for risk reduction efforts. This is achieved by analyzing natural hazards accidents data and using data science techniques to define data clusters and delve into the related factors affecting mortality. The climate-driven, natural disaster events from the International Disaster (EM-DAT) database have been thoroughly explored and visualized to obtain an overview of the current natural disaster situation. More specifically, this manuscript concerns the development of clustering algorithms and analytics to map fatalities and economic damage. The results of the analysis showed the extent to which climate change has a significant effect on resulting fatalities and economic losses from natural hazards accident scenarios. Besides the achieved results of this work, it is acknowledged how further studies should try to dynamically represent vulnerability as well as improve the quality and selection of integrated features to improve the representation of industrial aspects.