Creating JAABA classifiers.

Intraspecies aggression has profound ecological and evolutionary consequences, as recipients can suffer injuries, decreases in fitness, and become outcasts from social groups. Although animals implement diverse strategies to avoid hostile confrontations, the extent to which social influences affect escape tactics is unclear. Here, we used computational and machine-learning approaches to analyze complex behavioral interactions as mixed-sex groups of mice, Mus musculus, freely interacted. Mice displayed a rich repertoire of behaviors marked by changes in behavioral state, aggressive encounters, and mixed-sex interactions. A distinctive behavioral sequence consistently occurred after aggressive encounters, where males in submissive states quickly approached and transiently interacted with females immediately before the aggressor engaged with the same female. The behavioral sequences were also associated with substantially fewer physical altercations. Furthermore, the male’s behavioral state could be predicted by distinct features of the behavioral sequence, such as kinematics and the latency to and duration of male–female interactions. More broadly, our work revealed an ethologically relevant escape strategy influenced by the presence of females that may serve as a mechanism for de-escalating social conflict and preventing consequential reductions in fitness.