Classification of lameness in dairy cows using automatic milking system data and body condition score with machine learning

Lameness is a major welfare and productivity concern in dairy herds. This study investigated the influence of animal traits (parity, BCS) and environmental factors (farm, season) on locomotion score (LS) in lactating cows and assessed the impact of lameness on milking parameters using data from 3 Italian farms equipped with automatic milking systems (AMS). A total of 323 cows were evaluated biweekly for LS and BCS over 7 mo. The AMS data (n = 42,569 observations) were collected and analyzed with linear mixed models to assess relationships between LS and milking parameters. Multiple correspondence analysis was performed to explore variable associations, and a machine learning model (extreme gradient boosting) was trained to classify cows into 3 lameness classes. Cows in parity 3 or greater and thin cows showed significantly higher LS. Severely lame cows had reduced daily milk yield, fewer milkings per day, longer milking duration, and delayed milk flow, particularly in rear quarters. The machine learning algorithm, based on milking and cow-level features, achieved a balanced accuracy of 92% in classifying cows as nonlame, mildly lame, or severely lame. Shapley values revealed that BCS, parity, milk flow, and milking frequency were key predictive features. These findings confirm the potential of AMS and BCS data to support early detection of lameness. Integrating these data and machine learning offered an efficient approach to lameness monitoring without additional equipment.