Decoding individual dairy cow behavioral profiles through electronic feeder and drinker data

Individual animals can differ considerably in how they feed, drink, and compete for resources, yet much data analysis still relies on group-level comparisons that obscure these individual differences. Electronic feeding and drinking systems make it possible to characterize individual behavioral profiles within groups. Using 107 lactating Holstein dairy cows tracked over 207 d, we extracted 16 behavioral variables from electronic feed bin and water bin data, capturing conventional metrics (feeding and drinking intake, duration and meal frequency) alongside measures of non-nutritive visits, social competition (replacement events), and indicators of feed quality (e.g., proportion of fresh feed remaining upon arrival). Bayesian mixed-effects models quantified repeatability (between-individual variation in average behavioral expression) and predictability (between-individual variation in within-individual consistency). Variables capturing non-nutritive visits, social competition, and drinking duration showed the greatest and most consistent among-individual variation, while daily feed intake was among the least discriminating, suggesting individuals converge to similar intakes through different strategies. To identify behavioral profiles, we applied principal component analysis with varimax rotation to the 8 most repeatable and individually variable behavioral measures, then used k-means clustering on the rotated principal components. This yielded 4 distinct profiles: "frequently replaced" cows (the majority of the herd, frequently replaced at the bins), "contest-winning" cows (initiating many replacements with efficient, uninterrupted meals), "frequent drinker visitors" (many long drinking visits spread across the day), and "fragmented feeders" (many short, interrupted visits with frequent replacement events). These results illustrate how data streams from electronic feeders and drinkers can be used to build individual behavioral profiles that would not be apparent using generic group-level analyses. This framework may help researchers identify cows most susceptible to negative impacts when housed in competitive environments. To facilitate access across species and systems we developed moo4feed (https://www.skysheng.io/moo4feed/), an open-source R package designed to extract individual behavioral variables from electronic feeder and drinker data.