Fairness and Explainability of Random Forest Regression in Predicting Social Media Rescue Requests with Socio-Environmental Features: Case of Hurricane Harvey

Keywords: AI fairness, Random Forest, regression fairness, explainable AI, disaster rescue
Abstract Type: Virtual Paper Abstract

Abstract

AI fairness is tasked with evaluating and mitigating bias in algorithms that may discriminate towards protected groups. This paper examines if bias exists in AI algorithms used in disaster management and in what manner. We consider the 2017 Hurricane Harvey when flood victims in Houston resorted to social media to request for rescue. We evaluate fairness criteria (independence, separation, and sufficiency) of a Random Forest regression model trained to predict Twitter rescue request rates on social-environmental data. Nine sensitive attributes were considered: Social Vulnerability Index (SVI), each of its four themes (socioeconomic status, household composition and disability, minority and language, transportation and housing), and rates of access to any device, a smartphone, Internet, and cellular data plan. The seven significant predictors of rescue requests were percent of roads and of the area in the flood zone, wetland and forested cover, mean elevation, as well as rates of renter occupied housing and access to digital devices. Partial dependence plots linked higher environmental vulnerability, limited access to digital device, and low renter housing rates with higher rescue rate predictions. While the model satisfied the three criteria of fairness – independence, separation, and sufficiency – for most of the sensitive attributes, we found imperfect sufficiency based on the housing and transportation vulnerability themes. These results suggest that future AI modeling in disaster research could apply the same methodology used in this paper to evaluate fairness and help reduce unfair resource allocation that would deepen social and geographical disparities