Towards fairness-aware disaster informatics: An interdisciplinary perspective

Abstract

Collection of information from crowdsourced and traditional sensing techniques during a disaster offers opportunities to exploit this new data source to enhance situational awareness, relief, and rescue coordination, and impact assessment. The evolution of disaster/crisis informatics affords the capability to process multi-modal data and to implement analytics in support of disaster management tasks. Little is known, however, about fairness in disaster informatics and the extent to which this issue affects disaster response. Often ignored is whether existing data analytics approaches reflect the impact of communities with equality, especially the underserved communities (i.e., minorities, the elderly, and the poor). We argue that disaster informatics has not systematically identified fairness issues, and such gaps may cause issues in decision making for and coordination of disaster response and relief. Furthermore, the isolating siloed nature of the domains of fairness, machine learning, and disaster informatics prevents interchange between these pursuits. This paper bridges the knowledge gap by evaluating potential fairness issues in disaster informatics tasks based on existing disaster informatics approaches and fairness assessment criteria. Specifically, we identify potential fairness issues in disaster event detection and impact assessment tasks. We review existing approaches that address potential fairness issues by modifying the data, analytics, and outputs. Finally, this paper proposes an overarching fairness-aware disaster informatics framework to structure the workflow of mitigating fairness issues. This paper not only unveils both the ignored and essential aspects of fairness issues in disaster informatics approaches but also bridges the silos which prevent the understanding of fairness between disaster informatics researchers and machine-learning researchers.