A few weeks ago, we began piloting HunchLab 2.0 in Lincoln. HunchLab is a crime forecasting product from Azavea. HunchLab seeks to forecast crime by analyzing historical crime data and other information related to crime, such as weather, population density, and land use. HunchLab uses these data to predict the most likely areas for crime today, in eight hour blocks.
The output from HunchLab consists of a map of Lincoln containing a few dozen color-coded cells about the size of two square blocks. These are areas where the risk of crime is forecast to be heightened during the prediction period. These cells are approximately two blocks square. The colors represent the particular crime type for which the elevated crime risk is predicted.
HunchLab predictions are based on the analysis of Lincoln's crime data for the past five years. Where crime has occurred in the past provides a good clue as to where it will occur in the future. This is particularly true for recent crime, so crimes occurring in the immediate past are given more weight. Historical data also reveals patterns in time: crime has large peaks and valleys across the calendar and the clock.
Population density is an excellent predictor of many crime types, as is income. Densely packed low-income neighborhoods suffer more crime than sparsely populated suburban neighborhoods.
Land use and zoning impacts crime. Retail businesses, such as convenience stores, restaurants, hotels and motels, grocery stores, and so forth draw people together so the chance of offenders and victims encountering one another is increased. In addition, certain kinds of businesses are related to a significantly elevated risk of crime at or around the business, such as bars and liquor outlets.
Essentially, when you gather all these kinds of data together, you can make informed predictions about where crime is most likely to occur. All of the factors that go into the prediction are based on research about the causes and correlates of crime. In fact, the commercial product emerged from research conducted at Temple and Rutgers. HunchLab goes a step further, and tests the predictions: how well did the cells identified by HunchLab’s algorithm perform in predicting where the crime actually occurred in subsequent time periods? One of the distinguishing features of HunchLab is the ongoing testing of the model, and machine learning that adjusts the predictions on the fly.
HunchLab is far better at predicting crime locations than random distribution. Although Hunchlab is predicting crime at places where most seasoned Lincoln Police Officers would recognize higher risk from their experience, there are other places that aren’t so obvious. It is also worth noting that not all Lincoln police officers are equally "seasoned."
We have helped out with the development of HunchLab 2.0 by providing data and feedback, and we are one of a handful of agencies piloting the application. For a more detailed description of HunchLab under the hood, the underpinning in criminological theory, and the research upon which it is based, go here.
Predictive analytics are getting a lot of attention in policing these days, but the same techniques can be applied to fire and EMS work. I blogged about this a few years ago, when the concept was still pretty new. Here in Lincoln, it is possible to predict fairly accurately both where we'll be sending fire engines and medic units, and when those responses will be occurring. We are using this knowledge more than ever to make informed decisions about our operations at Lincoln Fire & Rescue.