Using machine learning to identify key predictors of maternal success in sheep for improved lamb survival

This study investigates key physiological, genetic, and environmental factors influencing maternal success in sheep to enhance lamb survival and maternal quality. Using data from native and crossbred prolific ewes in a high-altitude, cold-climate region, we applied machine learning models to predict mothering scores based on dam characteristics, birth conditions, and lamb attributes. Pregnant ewes were monitored 24 hours per day, beginning three days before parturition, with minimal human intervention. Predictor variables included dam breed, body weight, age, litter size, lamb genotype, lambing season, time of lambing, parturition duration, and lambing assistance. Several machine learning algorithms, including Random Forest, Decision Trees, Logistic Regression, and Support Vector Machines (SVM), were evaluated for predictive accuracy. The Random Forest model achieved the highest accuracy (67.2%) and demonstrated the best overall performance with a 0.41 Kappa statistic and the lowest mean absolute error (0.59). Feature importance analysis identified dam weight at birth, parturition duration, and lamb birth weight as the strongest predictors of maternal success. The Decision Tree model highlighted time of lambing, lamb genotype, and lambing assistance as key decision points for classifying mothering ability. Further analysis revealed that shorter parturition durations (≤ 38 min), unassisted lambing, and smaller litter sizes were associated with higher mothering scores. Breed-specific maternal differences were also observed, with crossbred prolific ewes exhibiting stronger maternal instincts. These findings provide actionable insights for precision livestock farming, emphasizing the importance of genetic selection, birthing management, and environmental monitoring to enhance maternal efficiency and lamb survival.