Behavioral selection for working dogs, particularly those used in olfactory detection, remains a major research focus. In a 2023 study published in Scientific Reports, Alexander W. Eyre and colleagues analyzed behavioral and environmental data from a cohort of 628 Labrador Retrievers raised by the U.S. Transportation Security Administration. The study used machine learning (ML) to predict which dogs would be accepted into formal training and which would be eliminated.
Three ML algorithms were applied to data collected across four time points over a 12-month foster period. The most accurate predictions occurred at the 12-month testing point (AUC = 0.68), where accepted versus eliminated outcomes were most distinguishable. However, all algorithms struggled to accurately classify eliminated dogs, possibly due to the complexity or variability of underperformance traits.
Feature selection techniques, including Principal Components Analysis and Recursive Feature Elimination, highlighted the significance of behavioral traits such as olfaction, possession, confidence, and initiative. These traits were especially prominent during specific tests, such as airport terminal search and retrieve tasks and environmental assessments.
The authors suggest that integrating behavioral testing with machine learning can improve future working dog selection protocols. They also emphasize the need for further research that includes not only cognitive and emotional attributes, but also social and environmental contexts—offering a more holistic approach to canine performance evaluation.
Source: Alexander W. Eyre, I. Zapata, E. Hare, J. Serpell, C. Otto, C. E. Alvarez. Scientific Reports, Volume 13, August 1, 2023.
