The Augurisk science team is excited to introduce our new crime risk assessment, available through our website and our free mobile app, Augurisk Now. This feature will help our users have a clearer picture on crimes happening in their block group, around their house, field, office, or favorite restaurant.
As we have been working on this for a few months now, we thought it would be interesting to share a bit more about our crime risk assessment. This assessment covers all Census Block Groups in the contiguous United States.
Looking for data is an important part of our job and is fundamental to any of the risk assessments we provide. Unfortunately, in this case, detailed crime data was only available in a few parts of the contiguous United States – and when it was, sometimes, it had a different data structure due to the way it was collected. Therefore, we decided to predict it with the help of artificial intelligence (AI). Let us explain.
First, we collaborated with a sociologist on identifying the factors that are considered substantial predictors for crime. Moreover, we also used the advantages of big data to analyze correlations between crime and thousands of variables characterizing a place. Ultimately, we selected 188 predictors, i.e. socio-economic, demographic, spatial, environmental, and law enforcement predictors. We also gathered local crime data collected and homogenized for 11 cities spread over the United States – thanks to the Crime Open Database (Ashby, 2018).
Afterwards, with the help of our data scientists and a machine learning expert, we developed multiple Machine Learning (ML) algorithms that were trained to predict crime based on the aforementioned 188 predictors, and selected the most performing one.
In the end, we were satisfied to notice that we successfully managed to reach prediction accuracy ranging between 73 and 77% in our tests (i.e. successful ‘guesses’ by our algorithm to predict whether a block group was in the first, second, third, and fourth quartiles, or in the 2 highest centiles of crime counts among more than 13,000 block groups in the U.S.).
For perspective, here is a view of our predictions for the number of yearly violent crimes, centered around New York City, NY:
Sources: Augurisk crime index, Basemap: OpenStreetMap Standard.
This model has been used to predict crime everywhere in the contiguous U.S., and you can now enjoy its results right from your phone, with our free App Augurisk Now, and in all your Augurisk projects, under the crime section.
We provide either observed or predicted data for the yearly occurrences of violent crime, property crime, motor vehicle theft, and vandalism. If you are travelling to unfamiliar places, Augurisk Now can send you alerts when you enter an area where crime has been identified as significant by our algorithms. If you are considering moving to a new home, the Augurisk website can provide you with our predictions in the crime section - along with our 10 other risk assessments at your location.
We are very proud of this update to our product, but before you go: keep in mind that the crime statistics we provide are predictions. As such, they can sometimes be imperfect. In particular, because our risk score is based on crime rates (per 100 000 residents), we tend to overestimate risk in areas where population density is low but traffic is high - such as urban parks, malls, airports, etc. As our model has been trained in urban contexts, it is also less precise in rural areas. Finally, we noted some overestimations in the number of crimes predicted in selected states, while our model produces satisfying results in relative terms. In other words, our crimes scores in % show a higher performance than our numbered predictions.
Crime is a complex social phenomenon, and we believe that our assessment can help you take more informed decisions. A scientific paper detailing our methodology has just been published in an international peer-reviewed academic journal. You can find all the details here.
Thanks for reading us and stay safe in these troubled times.