Can Unsupervised Machine Learning Models Be Used to Better Allocate Public Safety Resources?

Sergio Brilanti-Martinez and Dr. Brad Ray, School of Social Work, 42 W. Warren Ave. Detroit, MI 48202

The past summer has seen mass civil unrest in the United States as a response to dissatisfaction with existing public safety service models. Previous analysis of crime data using unsupervised machine learning methods is abundant. However, less research has been done to use these tools in the broader context of public safety and to guide areas of opportunity for public safety reform. This analysis will attempt to find patterns and discrepancies between public safety demand and supply in part by separating officer and citizen-initiated service calls for each type of crime or incident as well as incorporating complaint data. Of particular interest is examining any disparities among the percentage of calls of a type within a census track initiated by officer or citizen. Disparities in this variable among clusters can shed insights into behavioral patterns among both officers and citizens in how they respond to public safety concerns. Initial mapping of officer versus citizen-initiated service calls in ArcGIS Pro illustrates some types of service calls are initiated by officers at much higher rates in certain areas. Three datasets from the city of Detroit will be used with data from a 12-month period: 911 service calls, crime incidents, and police complaints. The data will be run through a k-means algorithm, the feature set will be filtered to get rid of noise, and the number of clusters will be determined using the elbow method. The modeling will be with the purpose of better understanding public safety demand and supply in the city of Detroit and of evaluating the usefulness of k-means modeling for determining public safety demand and supply and its potential role in generating actionable recommendations to key stakeholders.

Additional Abstract Information

Presenter: Sergio Brilanti-Martinez

Institution: Wayne State University

Type: Oral

Subject: Sociology

Status: Approved

Time and Location

Session: Oral 5
Date/Time: Tue 12:30pm-1:30pm
Session Number: 545
List other presenters in this same room and session