Decision tree for soldiers could reduce civilian deaths

Soldiers armed with three simple questions gain a lifesaving edge in distinguishing friends from suicide attackers, a new study finds.

This new decision-making formula can potentially reduce the number of civilians killed or injured while driving toward military checkpoints, convoys and ground patrols, say psychologists Niklas Keller of Charité University Hospital and Konstantinos Katsikopoulos of the Max Planck Institute for Human Development, both in Berlin.

The researchers analyzed 1,053 encounters between NATO soldiers and nonhostile vehicles in Afghanistan from 2004 through 2009. The new decision-making approach would have reduced civilian casualties in those incidences from 204 to 78, the researchers conclude July 15 in the European Journal of Operational Research. NATO data for testing the decision strategy was obtained from WikiLeaks, a group that publishes a variety of leaked documents.

The new question format shows promise as a way to save civilian lives in war zones, says operations research professor Raimo Hämäläinen of Aalto University, Finland.

Keller and Katsikopoulos developed a decision tree based on a review of military literature and interviews with eight German military personnel who had experience manning checkpoints in Kosovo or fighting in Afghanistan. The tree has soldiers first ask whether an approaching vehicle contains more than one person. If so, it’s deemed safe. In 84 percent of cases, that question was enough to identify truly nonhostile vehicles.

The power of first checking a vehicle’s occupancy was reflected in the fact that all seven successful suicide attacks on NATO forces in Afghanistan during the study period involved single-person vehicles.

If a one-person vehicle is approaching, the tree’s next question asks whether that vehicle complies with requests to slow down or stop. If not, it’s labeled hostile. Otherwise, a final question asks whether “there are no further threat attributes,” such as an approaching car displaying a color or being of a make identified in intelligence reports as belonging to suicide attackers. If any such threats are seen, the car is treated as hostile.

The new tree isn’t perfect. It would have reduced civilian casualties for multiple-occupant vehicles – always deemed safe in the new formula – by 183.  That’s how many were killed or injured by soldiers despite the fact that none of those vehicles contained attackers. On the negative side, while soldiers harmed 10 civilians in single-occupant, noncompliant vehicles, that figure would have reached 74 by using the decision tree, the researchers say.

During the time period analyzed, soldiers also harmed eight civilians in single-occupant, compliant vehicles and three of four civilians in single-occupant, compliant vehicles with threat features, making a total of 204 civilian casualties. The decision tree would have classified no one as hostile in the single-occupant, compliant vehicles and all four civilians as hostile in single-occupant, compliant vehicles with threat features, adding up to a total of 78 potential civilian casualties using the strategy.

Keller wants to see how the decision tree — which can be modified if attackers change tactics — works in real-world situations. “If this was a drug, it would be time to do a clinical trial,” he says.