AI For Bias Detection: Investigating the Existence of Racial Bias in Police Killings

A recent paper in the Journal of Politics applied machine learning models to analyze racial disparities in police shooting fatalities. It suggested building generalizable binary classifiers simply based on those social and demographic factors. This recent research in 2019 was not able to establish a convincing machine learning model account for the observed racial bias in fatalities resulting from police shootings. This paper proposes key improvements made from the previous machine learning research through the use of class weights, random oversampling, and a more expansive dataset. We are able to show generalizable relationships between factors related to a police killing and the race of the fatality with over 80 percent accuracy based on the US national police shooting fatality data in recent years. This research suggests areas for further investigations of racial bias in police killings with supervised machine learning as a methodology for algorithmic and data bias detection.