Temporal Analysis of World Disaster Risk:A Machine Learning Approach to Cluster Dynamics

TL;DR

This study analyzes global disaster risk dynamics from 2011 to 2021 using the World Risk Index, revealing persistent high susceptibility clusters and limited effectiveness of current policies, emphasizing the need for innovative risk management strategies.

Contribution

It introduces a semi-supervised clustering approach with high accuracy and evaluates the predictive power of supervised learning models for disaster risk, highlighting policy shortcomings.

Findings

Global risk clusters remain stable over time.

Logistic regression predicts risk clusters with nearly 99% accuracy.

Low probability (1%) of countries shifting to safer risk levels within five years.

Abstract

he evaluation of the impact of actions undertaken is essential in management. This paper assesses the impact of efforts considered to mitigate risk and create safe environments on a global scale. We measure this impact by looking at the probability of improvement over a specific short period of time. Using the World Risk Index, we conduct a temporal analysis of global disaster risk dynamics from 2011 to 2021. This temporal exploration through the lens of the World Risk Index provides insights into the complex dynamics of disaster risk. We found that, despite sustained efforts, the global landscape remains divided into two main clusters: high susceptibility and moderate susceptibility, regardless of geographical location. This clustering was achieved using a semi-supervised approach through the Label Spreading algorithm, with 98% accuracy. We also found that the prediction of clusters achieved through supervised learning on the period considered in this study (one, three, and five years) showed that the Logistic regression (almost 99% at each stage) performed better than other classifiers. This suggests that the current policies and mechanisms are not effective in helping countries move from a hazardous position to a safer one during the period considered. In fact, statistical projections using a scenario analysis indicate that there is only a 1% chance of such a shift occurring within a five-year timeframe. This sobering reality highlights the need for a paradigm shift. Traditional long-term disaster management strategies are not effective for countries that are highly vulnerable. Our findings indicate the need for an innovative approach that is tailored to the specific vulnerabilities of these nations. As the threat of vulnerability persists, our research calls for the development of new strategies that can effectively address the ongoing challenges of disaster risk management