Deliverables

  • A detailed statistical analysis of the relationship between street light outages and crime incidents in San Francisco
  • A list of San Francisco neighborhoods with a statistically significant relationship between crime and street light outages
  • A script to calculate weighted crime rate, which can be used for any city that has data on street light outages and crime incidents organized by date and latitude/longitude points
  • A model for optimization of street light repair that can be continuously updated with new data, as crime rate and street light functionality change perpetually. It may also be used for street light repair prioritization in other cities or for city repairs other than street light outages.

Discussion

Only 10 of the 120 neighborhoods in San Francisco have been analyzed. In the future, a statistical analysis and traveling salesman problem would need to be completed for each neighborhood so that they may be incorporated in the optimization problem. Because our Python code (used to calculate the weighted crime rates of street lights) has a run time of about 1 hour, we were unable to expand our analysis further. If further analysis is desired, this task may be completed by Noodle.ai who has access to greater computing power and general computer science expertise.

Possible Improvements

First, we could make our calculation of weighted crime rates more accurate by filtering out all duplicate repair tickets and crime incident reports. For the purpose of our project, we simply held the assumption that duplicated tickets and reports would cancel out due to random distribution.

Second, we could improve our method to route street light repairs in neighborhoods. By incorporating exact roads and traffic patterns through the use of the Google Maps API, we would be able to provide an exact route that a repairman should take in each neighborhood. 

Last, we could research and implement alternative operation research techniques for our final optimization problem. Ideally, the program would result in a prioritized list, instead of needing to change the constraints to determine prioritization.

Further Questions

  • Why do certain neighborhoods have a higher correlation between street light outages and crime rate?
  • What causes this correlation, and why does it differ between neighborhoods?