A Machine Learning Approach to Analyzing Officer-involved Shootings in the United States

• In Event: Law Enforcement, Shootings, and Guns

Abstract

This paper examines officer involved-shootings that have occurred between 2000 and 2015 in the United States. Researchers retrieved their data from SHOT (Statistics Help Officer Training) database, which contains more than 3000 shootings with more than 50 variables. They characterize them by a binary data matrix, where rows in this matrix represent individual shootings. Then, a non-negative matrix factorization technique was used to identify shooting patterns that best summarize this data matrix. The statistical test to assess similarities and dissimilarities discovered some shooting patterns across four main regions of the U.S.: Northeast, West, South, and Midwest. Research confirmed that there are some interesting shooting patterns in terms of subject, officer and incidental variables in each aforementioned region.

Authors

• Hasan T. Arslan, Pace University

• Murat Dundar, Indiana University-Purdue University, Indianapolis