Skip to content

Building a simple predictive policing model to demonstrate the inherent biases incident in commercial systems

Notifications You must be signed in to change notification settings

vaibhavram/pred-pol

 
 

Repository files navigation

Model Analysis: To Predict and serve?

This repository contains code, data and tests to analyze the PredPol system presented in "To Predict and Serve?," published in volume 13 issue 5 of Significance, and authored by Kristian Lum and William Isaac of the Human Rights Data Analysis Group (HRDAG). The original repository which contains the replication materials for the paper can be found in HRDAG/predictive-policing.

Our analysis is entirely contained in the analyses/ folder.

Our model uses past crime data in individual geographical "bins" weighted by an exponential decay kernel to compute bin scores, and we would use these to determine the likelihood of crimes for the next day.

Data

  • drug_crimes_with_bins.csv — Data collected by OpenOakland.
  • bin_touching_dictionary.rds — Bin numbers mapped to a list of neighboring bin numbers.
  • lum_testing_rates.rds — Capture rates for Kristian Lum's model.
  • parameter_tuning.rds — Capture rates for our model with different parameters.

Code

  • auc.R — Generates an ROC curve to compare results.
  • model.R — Contains code for our model.
  • paramTuning.R — Uses model.R to tune its parameters.
  • resultGenerate.R — Plots our accuracy against Kristen's accuracy for sampled dates.

About

Building a simple predictive policing model to demonstrate the inherent biases incident in commercial systems

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 50.0%
  • Jupyter Notebook 38.1%
  • R 9.1%
  • Makefile 2.8%