Promising Results from Standardizing Police Activity Data

• In Event: Poster Session 2

Abstract

It is estimated that only 20-40% of police activity is related to responding to crime with noncriminal quality of life and disorder issues occupying most police activity. An agency’s computer-aided dispatch system captures most of the data to complete this picture, however, there are no existing standards or agreed-upon approaches to capturing CAD data. Across jurisdictions, unstandardized CAD data limits regional, state, or national examinations of police activity.

Recognizing the need for better data about police activity and community need, the Bureau of Justice Statistics conceptualized a partnership with RTI to develop a hierarchical taxonomy along with a machine learning model to standardize raw CAD data from disparate law enforcement agencies. Using the latest advances in large language models and artificial intelligence approaches, the BJS/RTI team has developed a text classifier with data from over 60 agencies that ingests, cleans, and standardizes raw agency data allowing direct comparison across agencies and further trend analysis on police activity at a regional level. We created a two-level classification model which is coupled with a user interface that allows law enforcement agencies and other users to easily run their own CAD data through our text classifier to standardize their activity into our taxonomy.

Authors

• Sean E. Wire, RTI International

• Christopher Williams, RTI International

• Anthony Berghammer, RTI International

• Andrea Gardner, Bureau of Justice Statistics, US DOJ

• Jacob Cramer, RTI International