Follow Datanami:
October 12, 2015

Mining Military Data to Predict Violent Offenders

Researchers at Harvard University’s medical school and other institutions studying post-traumatic stress disorder and other mental health issues plaguing American war veterans report they have developed an “actuarial model” based on machine learning that can help predict future violent crimes by U.S. soldiers.

The researchers noted that their model based on an administrative dataset of more than 975,000 U.S. soldiers could help pinpoint those most prone to violent crime. That information could then be used to make existing interventions more effective.

A U.S. Army database was created between 2004 and 2009 to study patterns in returning soldiers called the “Army Study to Assess Risk and Resilience in Servicemembers.” The database revealed that 5,771 returning soldiers committed violent crimes during the study period, ranging from murder and aggravated assault to arson and robbery.

Investigators combined these data along with other administrative records such as military career and demographic information along with crime and medical records to build the actuarial model. The model applied machine-learning methods to the reported crimes among male and female service members.

The investigators said in a paper published by Cambridge Journals in Psychological Medicine that they validated their model in an independent sample spanning 2011 to 2103.

“Key predictors were indicators of disadvantaged social/socioeconomic status, early career stage, prior crime, and mental disorder treatment,” the researchers reported. Of all crimes committed by returning veterans, more than one-third were committed by soldiers in the highest risk group predicted by the actuarial model. The total jumped to more than half for later validation sample, the researchers reported.”

“Although the results suggest that the models could be used to target soldiers at high risk of violent crime perpetration for preventive interventions, final implementation decisions would require further validation and weighing of predicted effectiveness against intervention costs and competing risks,” the researchers concluded.

Along with Harvard Medical School, researchers from the University of Virginia School of Law along with other university and Veteran Administration hospitals participated in the analytics study.

According to a report in the Los Angeles Times, the researchers used 38 databases to build their model. The data sets contained 446 variables for each U.S. soldier who served between 2004 and 2009.

The tool gives the Army the ability “to identify high-risk soldiers without carrying out expensive one-on-one clinical assessments,” Harvard psychologist Anthony Rosellini, the study’s lead author, told the Times.

Researchers earlier used the same method to create an analytics tool for identifying soldiers at greatest risk of suicide. Suicides by returning U.S. veterans have reached historically high numbers over the last five years, according to Defense Department statistics released earlier this year, reaching a reported 288 active-duty personnel in 2014. The total includes soldiers, sailors and airmen.

Recent items: 

How DARPA Does Big Data

Army Deploying on Big Data Battlefield

Datanami