Machine learning-based prediction of pre-weaning lamb survival using animal-, housing-, and management-related factors

The aim of the study is to classify lambs that will survive or die before weaning by using animal- and farm-related factors as predictors through machine learning (ML) algorithms, and to identify potential risk factors associated with lamb mortality in housed management systems. Survival records from birth to weaning of a total of 5539 lambs were analysed from ten farms, which reared Kivircik sheep. To predict whether the lambs will survive from birth to weaning, Decision Tree (DT), Random Forest (RF), K-Nearest Neighbors (KNN), Naïve Bayes (NB), Boosting, Logistic Regression (LR), Support Vector Machine (SVM) and Multilayer Perceptron (MLP) classifiers were tested. The survival of lambs in the raw dataset was 91.7 %. The highest accuracy (0.925) in classifying living or dead lambs was obtained by Boosting algorithm, while the second highest performance (accuracy of 0.897) was shown by RF. NB, LR, and SVM algorithms achieved relatively lower classification accuracies, ranging between 65 % and 67 %. According to the Boosting algorithm, birth weight was identified as the variable with the highest relative influence with 43.4 %. It was followed by birth month (12.2 %), number of ewes per shepherd (11.2 %), floor space per ewe (10.0 %), and birth rank group (7.0 %). In conclusion, Boosting algorithm demonstrated high classification accuracy in predicting lamb survival. Moreover, the strong predictive influences of birth weight, number of ewes per shepherd, floor space per ewe, and birth rank group indicate the importance of focusing on gestational nutritional management, husbandry conditions, and overall herd management practices in developing intervention strategies to reduce lamb mortality in housed management systems. • Eight machine learning algorithms were used to predict lamb survival status. • The survival of lambs in the raw dataset was 91.73 %. • The highest accuracy (0.925) was obtained with Boosting algorithm. • Birth weight was the variable with the highest relative influence with at 43.4 % • Birth month, number of ewes per shepherd, floor space per ewe were high influence