An iterative approach to identify key predictive features of fear reactivity and fearfulness in horses (Equus caballus).

This study extends previous findings by applying artificial intelligence (AI) methods to a larger dataset to identify key features that predict fear reactivity (i.e., immediate reaction to fear inducing stimuli) and fearfulness (i.e., a stable personality trait) in 101 Lipizzan horses. The analysis included 221 morphological, kinematic, behavioral and management measurements per horse. Previous findings were confirmed, as body and head size were identified as promising predictors of aspects of fear-related trait. Using an iterative AI approach, six key features for fear reactivity and nine for fearfulness were identified, with decision tree analysis highlighting significant features that were relevant for equal or more than 10 horses. A 96% behavioral overlap between reactivity and fearfulness was observed, indicating a strong correlation. However, key predictive features differed between the two traits, with correlation coefficients not exceeding 0.57. This study highlights the complexity of fear-related traits and emphasizes that specific phenotypes more accurately predict reactivity and personality in adult horses when AI methods are used. These methods may provide objective, data-driven insights into horses' behavior, which could support more informed and individualized decisions in management, training and breeding.