2026 Conference of Research Workers in Animal Diseases46

CRWAD 2026	 ABSTRACTS

12 - Integration of sensor data and lesion history for predictive modeling of dairy cow lameness.

T. Rahman*, E. Shepley, and G. Cramer,

University of Minnesota, Twin Cities, MN

Timely identification of individual lameness is critical for dairy herd health and productivity. While traditional methods 
rely on subjective visual scoring, the adoption of precision dairy technologies provides an opportunity for objective, 
continuous monitoring. The study’s objective is to develop and validate a predictive model for locomotion score 
using behavioral data collected from neck-mounted sensors and historical foot lesion occurrence. Sensor data were 
collected from 2 commercial dairy farms in Minnesota between January and July 2025. The dataset included daily 
totals for eating time, rumination time, and inactive time, measured by collar sensors, alongside corresponding visual 
locomotion scores (0–3) provided by experts (score 0 is a non-lame and score 3 is a severely lame cow). After data 
cleaning, a total of 6,561 cow-days from 846 unique cows were available for analysis. The locomotion scores were 
categorized as lame (score 0 and 1; n = 176) and non-lame (score 2 and 3; n = 670) cows. To account for repeated 
measures on individual cows, the dataset was split into training and testing sets by individual cow, ensuring no over-
lap between sets. A random forest model was developed using average eating, rumination, and inactive time from 
the 2 weeks before the locomotion score date, their standard deviations, days in milk, lactation number, and lesion 
history as predictors. Model performance was evaluated across 30 randomized iterations. While the predictive model 
achieved high average accuracy (76%), its performance was driven by high specificity (84%) at the expense of low 
sensitivity (46%), indicating the model was proficient at identifying non-lame cows but struggled to detect lame cows. 
Lower sensitivity might be caused by unbalanced classes in the dataset. The initial results indicate that behavioral 
measures can explain a substantial portion of the variation in locomotion scores.

Key Words: lameness, dairy cattle, wearable sensor, machine learning, animal behavior, precision dairy