Characterizing dairy cows based on their daily behavior deviation patterns relative to the herd using a clustering approach.

Abstract

Large variability in the behavioral patterns among individual cows has been reported in previous studies, which compromises the interpretation of herd behavioral patterns. Assessing the behavioral variability of cows at the subgroup level may make the interpretation clearer. The aims of the current study were (1) to assess the individual behavioral variability among cows by clustering individuals based on deviations from herd level patterns in lying duration, number of steps, and number of posture changes at the daily level and during specific time periods within a day, separately for periods without and with pasture access, and (2) to characterize the identified clusters according to their behavioral variables and parity. Accelerometer data, including lying duration, number of steps, and posture changes in 15-min intervals of 185 individual cows in the year 2019, were retrieved from one commercial dairy farm in the Netherlands. From the behavioral data, 13 variables describing the behavioral deviation from the herd level in lying duration, number of steps, and posture changes were extracted per individual cow, including absolute deviation across a day and signed deviation in specific time periods within a day. K-means clustering based on the extracted variables was performed to cluster individual cows in periods without pasture access and with pasture access separately. In both clustering analyses, the same 4 types of individual cow clusters were identified. The behavioral deviation patterns of the 4 types of clusters were further characterized based on the mean values of variables of individual cows per cluster: 1) locomotive; 2) standing; 3) restless; 4) lying. Cows in the locomotive cluster had shorter lying duration and higher number of steps than the herd level. Cows in the standing cluster had shorter lying duration, lower number of steps, and fewer posture changes than the herd level. Cows in the restless type cluster had higher lying duration, number of steps, and more posture changes than the herd level. Cows in the lying cluster had higher lying duration and lower number of steps than the herd level. 72.4% of the individual cows were consistently clustered in the same type of cluster, independent of whether cows were on pasture or not. The majority of cows in parity 1 were clustered in the locomotive type of clusters. Most cows in parity 2 were clustered in the standing type of clusters, whereas most cows in parity 3 or greater (parity 3⁺) were clustered in the lying type of clusters. In conclusion, 4 distinct types of behavioral deviation patterns were observed among individual cows via a clustering approach and were generally consistent in periods without and with pasture access. In general, the cows in parity 2 and parity 3⁺ had lower number of steps and posture changes than cows in parity 1. This study demonstrates the potential to use clustering to improve interpretation of individual behavioral variability and patterns.