A new effort called the Durkheim Project aims to predict which military veterans are at the highest risk of suicide by monitoring their posts to social media. The suicide rate among veterans has climbed to 22 per day, according to the Department of Veterans’ Affairs, while a record 349 active-duty soldiers committed suicide in 2012.
The team behind the Durkheim Project thinks it can help curb this problem by using machine learning across a big database in order to score veterans’ suicide risk in real time. Patterns and Predictions, a New Hampshire-based company with ties to Dartmouth University, is spearheading the effort, which relies heavily on the company’s software for analyzing unstructured linguistic data. The Defense Advanced Research Project Agency, or DARPA, provided funding for the project, and Dartmouth with big data vendors Attivio and Cloudera providing technological support.
The project has actually been around since 2011, although the first two years were spent building the platform for collecting, storing and analyzing the data (that’s where Attivio and Cloudera came in) and training the predictive models based on suicide data from the Department of Veterans Affairs. Patterns and Predictions claims it and its co-researchers from Dartmouth and the Department of Veterans Affairs were able to accurately assess suicide risk 65 percent of the time using these models.
In theory, accuracy should improve as the team collects more data — especially social data — from veterans who sign up to be part of the project. Facebook is on board as a partner, and the project also monitors activity on Twitter and Google+.
There is, however, one big caveat to the project: At this point, it’s only a study, which means researchers are not authorized to act, even if they see someone increasingly at risk of committing suicide. Once there’s an accurate enough understanding of the real factors contributing to suicide risk, it appears the goal is ultimately to provide doctors a dashboard through which they can monitor their patients’ statuses in near real time.
The Durkheim Project is one of a host of efforts right now trying to take advantage of new data sources, better technologies and advances in predictive analytics in order to effect positive change. We’ve covered a handful already, including ongoing efforts by organizations like the SumAll Foundation and DataKind to help non-profit agencies get a better handle on their data, and a Google-spearheaded fight against human trafficking. Researchers are especially keen on social media data, using it to study everything from the spread of disease to bullying.
I only hope that all this effort isn’t wasted on prediction without ever looking into underlying causes. The issue of veteran suicide, in particular, reminds me of Quid founder Sean Gourley’s talk at Structure: Data about data science versus data intelligence, with the former being about prediction and the latter being about solving problems. Whether it’s suicide, gun violence or Gourley’s work to predict attacks during wars, being able to predict events is great and necessary, but it ends up just being a perpetual Band-Aid solution if we never actually study the deeper, dirtier data that might help us figure out why people do things and attack issues at that level.
Feature image courtesy of Shutterstock user Rob Hainer.