Human-automation interaction in a simulated cabin baggage screening task with automated explosives detection

Abstract

This study investigated human-automation interaction with automated explosives detection systems for cabin baggage screening (EDSCB) depending on automation reliability. We conducted a laboratory experiment with 118 screeners of an international airport, which performed in a simulated X-ray screening task. We tested human-machine system performance in four test conditions: one baseline group with no EDSCB and three practically relevant EDSCB algorithms that varied systematically in automation reliability measures, i.e. accuracy, d’ and positive predictive value (PPV). Screeners had to detect improvised explosive devices, bare explosives, guns, or knives. EDSCB increased human-machine system performance for detecting explosives due to direct cueing of targets. The benefit of EDSCB depended on target difficulty and automation reliability in terms of d'. High false alarm rates caused a cry-wolf effect, especially for bare explosives. PPV of EDSCB explained screeners' compliance best. Our results suggest that operators use a probability-matching decision strategy, in particular for difficult targets