Porcine Facial Inference: A Computational Assessment of Facial Biometrics as Predictors of Conspecific Aggression in Swine — ASN Events

Porcine Facial Inference: A Computational Assessment of Facial Biometrics as Predictors of Conspecific Aggression in Swine (#640)

Catherine G McVey 1
  1. Animal Science + Statistics, North Carolina State University, Raleigh, NORTH CAROLINA, United States

Within the equestrian community, there is a considerable amount of antiquated knowledge relating variations in the structural features of a horse’s face to aspects of innate personality. Facial Width-to-Height ratios have been correlated with aggression in capuchin monkeys, but an analogous relationship has not been explored in a porcine model. The purpose of this study was to assess the efficacy of facial biometrics in the prediction of individual differences in the innate aggressive tendencies of swine. To do this, facial photographs were acquired from 120 piglets at 24-48 hours of age. Algorithms developed using the image processing tools in MATLAB for a previous project with horses were adapted to porcine facial structures, and a total of 38 facial biometrics were extracted from each piglet image. At two weeks of age, prior to weaning and mixing, 46 of these piglets were then randomly paired by gender and subjected to a paired encounter test with an unfamiliar conspecific in a novel environment. From behaviors observed during this encounter, piglets were assigned binary classifications for proactive and reactive aggression. Wilcoxon ranked sum tests were used to screen for facial metrics with categorical potential for either response, which were then added to a series of increasingly complex logistic regression models optimized using the R software package. The current measures considered by the industry to inform grouping decisions - gender and weight - did not yield statistically significant models. Addition of facial biometrics significantly improved the predictive potential of both models (p<0.001), but only yielded a strong R2 value (0.50) for the reactive aggression model. Addition of interaction terms with fitness measures and facial metrics of the opponent were needed to achieve greater accuracy with the proactive aggression model. Final models for both proactive and reactive aggression yielded strong R2 values of 0.64 and 0.69 respectively.

  1. Camerlink, Irene. “Sociable Swine: Indirect Genetic Effects on Growth Rate and their Effect on Behavior and Production of Pigs in Different Environments.” Diss. Wageningen University, Wageningen. 2014.
  2. Carre, Justin et al. “In Your Face: facial metrics predict aggressive behavior in the laboratory and in professional hockey players.” Proc R. Soc. B 275-2008 pp. 2651-2656.
  3. Carre, Justin et al. “What Lies Beneath the Face of Aggression.” SCAN 8-2013 pp. 224-229.
  4. Lefevre, Carmen et al. "Facial Width-To-Height Ratio Relates to Alpha Status and Assertive Personality in Capuchin Monkeys." PLoS ONE 9(4). 2014.
  5. McVey, Catherine. “Image Analysis for Determining the Characteristics of Animals.” US Patent 20120288160. May, 2012.
  6. Tellington-Jones, Linda. “Getting in TTouch: Understanding and Influencing Your Horse’s Personality.” Trafalgar Square Publishing. Vermont. 1995
  7. Trut, Lyudmila N. “Early Canid Domestication: The Fox-Farm Experiment.” American Scientist, Vol. 87 March-April 1999, pp. 165-169.
  8. Turner, Simon. “Breeding Against Harmful Social Behaviors in Pigs and Chickens: State of the Art and the Way Forward.” Journal of Applied Behavioral Science. Vol. 134. July, 2011.
  9. Wong, Elaine et al. “A Face Only An Investor Could Love: CEO’s Facial Structure Predicts Their Firms Financial Performance” Psychological Science, Vol. 22 No. 12. December 2011. pp 1478-1483.