Behavioral components define operational suitability metric for detection dog success

To ensure effective development and assessment of detection dogs, Auburn University Canine Performance Sciences program has implemented a structured behavioral evaluation at four developmental milestones.

At each timepoint, trainers scored candidate detection dogs on 11 subtests evaluating detection performance and environmental soundness using behavioral component (BC) scores. Additionally, trainers rated each dog's overall suitability for operational detection work using a singular global score. To investigate the relative importance of different metrics in predicting final outcomes, we leveraged a longitudinal dataset collected from 304 dogs.

Using a machine learning classifier, we discovered that suitability (S) scores at 12-months of age were highly predictive of training outcomes, achieving a prediction accuracy of 74.3%. The S scores appear to reflect a synthesis of subtest scores across relevant behavioral domains, illustrated by their correlative structures. To determine if S scores could be replicated using only BC scores, we estimated computationally synthesized suitability (CSS) scores at each timepoint using five significant BC scores as predictors in a linear model. Strong correlations were observed between trainer-reported suitability and CSS, with spearman's correlation coefficients exceeding 0.84 at all timepoints, demonstrating CSS scores' excellent ability to predict suitability. As predictors of training outcome, trainer-reported S scores demonstrated high accuracy, while CSS scores exhibited high precision (76.7%). Notably, both suitability metrics increase over time in dogs deemed operationally capable but not in unsuccessful dogs.

Our findings provide novel insights into the importance of data-driven approaches by combining expert trainer global assessments of suitability and BC scores with computational modeling in detection dog development.