MIDS Capstone Project Spring 2019

OptiPol: Discrimination-Aware Predictive Policing And Resource Optimization

Police departments have lately been struggling with a diminishing number of resources and a growing number of complaints of police discriminative behaviour. Current predictive policing solutions in the market generally work by predicting crimes using solely the historical crime data, which may already have some discrimination bias embeded into them due to police behavior in the past.

Optipol tries to leverage other datasources along with the historical crime data while removing possibly biased features to train less-biased predictive models. It also tries to create awareness to possible discrimination by showing a Fairness Indicator along with other KPIs which is constantly measuring the bias in crime predictions and patrol deployment plans.

Lastly, Optipol helps police departments plan faster by offering a patrol deployment planning area integrated with crime predictions where departments can plan deployments with real-time visualization of KPIs such as coverage of crimes by deployed units, the distance units will have to drive to reach their assigned destination and even how discriminative this deployment is. Through the use of mathematical optimization techniques, Optipol is also able to automatically generate deployment plans which tries to maximize the coverage in the entire city, while minimizing the distance patrols have to drive and avoiding underpolicing or overpolicing areas of the city. It can also create this deployment while maximizing the fairness of deployment by automatically reducing discrimination in the deployment.

Optipol is entirely built on a microservices architecture, which allows you to scale up or down individual parts of the solution as required. It also provides user security and authentication out-of-the-box.

The full list of features of Optipol:

  • Awareness of discrimination in machine learning results through indicators along with the predictions
  • Additional features besides historical crime data offered for reducing bias in predictive models
  • Re-train or try new predictive models by selecting desired features and algorithms and switch between models on-the-fly all through Optipol's interface
  • Plan unit deployments through Optipol's planning interface, with real-time calculation of KPIs
  • Use mathematical optimization to get the best possible deployment plan in seconds
  • User authentication and security built into the product out-of-the-box
  • Microservices architecture allows scaling individual pieces of the solution up or down as require
Optipol: To Predict and Protect
Optipol Logo
Crime Prediction Dashboard
Crime Predictions Dashboard
Deployment Planning Area
Deployment Planning
Machine Learning Training and Model Switching
Machine Learning Training and Model Switching

Last updated:

October 1, 2019