Machine learning prediction and classification of behavioral selection in a canine olfactory detection program.

Area of Science:

• Canine behavioral science
• Machine learning applications
• Animal-assisted interventions

Background:

• Growing interest in canine behavioral research for working dogs.
• Utilized a dataset of 628 Labrador Retrievers from a Transportation Safety Administration olfactory detection cohort.
• Data collected over a 12-month foster period at four time points.

Purpose of the Study:

• To perform Machine Learning (ML) prediction and classification studies of behavioral traits and environmental effects in working dogs.
• Identify key behavioral traits and environmental factors crucial for olfactory detection dog selection.
• Guide future research on cognitive, emotional, social, and environmental influences on working dog performance.

Main Methods:

• Applied three supervised ML algorithms for prediction and classification.
• Utilized Principal Components Analysis and Recursive Feature Elimination with Cross-Validation for feature importance.
• Analyzed data from four time points over a 12-month period.

Main Results:

• ML algorithms robustly predicted acceptance into training programs.
• Distinguishing eliminated dogs showed poorer performance (approx. 25% accuracy).
• The 12-month testing point offered the best predictive ability (AUC = 0.68).
• Olfaction and possession traits were key for airport search tests.
• Possession, confidence, and initiative traits were important for environmental tests.

Conclusions:

• Identified critical behavioral traits (olfaction, possession, confidence, initiative) for olfactory detection dog selection.
• Highlighted the importance of specific tests, environments, and the 12-month time point for effective selection.
• Demonstrated the utility of ML in optimizing working dog selection processes.