Machine learning algorithms for lamb survival

Highlights

• Lamb survival is influenced by the culmination of a sequence of interrelated events.

• MLA offer great flexibility for complex interactions among variables.

• Grooming behaviour is the first determinant for mothering ability.

• SMO was found best in predicting lamb behaviour.

Abstract

Lamb survival is influenced by the culmination of a sequence of often interrelated events including genetics, physiology, behaviour and nutrition, with the environment providing an overarching complication. Machine learning algorithms offer great flexibility with regard to problems of complex interactions among variables. The objective of this study was to use machine learning algorithms to identify factors affecting the lamb survival in high altitudes and cold climates. Lambing records were obtained from three native breed of sheep (Awassi = 50, Morkaraman = 50, Tuj = 50) managed in semi intensive systems. The data set included 193 spring born lambs out of which 106 lambs were sired by indigenous rams (n = 10), and 87 lambs were sired by Romanov Rams (n = 10).

Factors included were dam body weight at lambing, age of dam, litter size at birth, maternal and lamb behaviors, and lamb sex. Individual and cohort data were combined into an original dataset containing 1351 event records from 193 individual lambs and 750 event records from 150 individual ewes. Classification algorithms applied for lamb survival were Bayesian Methods, Artificial Neural Networks, Support Vector Machine and Decision Trees. Variables were categorized for lamb survival, lamb behavior, and mothering ability. RandomForest performed very well in their classification of the mothering ability while SMO was found best in predicting lamb behavior. REPtree tree visualization showed that grooming behavior is the first determinant for mothering ability. Classification Trees performed best in lamb survival. Our results showed that Classification Trees clearly outperform others in all traits included in this study.

Introduction

Lamb survival is a complex trait influenced by many different factors associated with management, climate, behavior of the ewe and lamb, and other environmental effects (Tomaszyk et al., 2014, Aktaş et al., 2015; and Moraes et al., 2016). Brien et al. (2010) suggested selecting related traits that are more reliably evaluated to make genetic improvement in lamb survival rather than improving it through genetic selection since heritability estimates of lamb survival are typically low (0.00–0.11; Safari et al., 2005). Correlated traits (recorded at lamb tagging) include birthweight (BWT), birth coat score (BCS), maternal behaviour score (MBS), lamb ease (LE), rectal temperature (RT), visually assessed lamb vigour (OBV), five timed lamb behaviours and three skeletal measures, crown–rump length (CRL), metacarpal bone length (ML) and thorax circumference (THO), (Fogarty et al., 2007).

Increased litter size is one of the biggest contributors to higher profits on lamb production. Crossbreeding with prolific sheep breeds is a way of increasing proportion of ewes having twins and triplets. However, lamb survival is an important issue in high litter size for sheep flocks. Davis et al. (1983) reported that as mean litter size increases above 1.7, the decline in single-bearing ewes is offset by an increase in triplet-bearing ewes. In studies investigating the survivability of lambs from mixed-age ewes, it was found that the lamb’s birth weight is a strong driver of lamb survival (Yapi et al., 1990; Morel et al., 2008); and researchers reported that lambs weighing <3 kg at birth have a lower survival rate from birth to weaning (Nowak and Lindsay, 1992). Neonate survival is dependent on the coordinated expression of appropriate behaviors from both mother and lamb (Dwyer, 2003); and behavioral interactions are much more important for prolific sheep with higher litter size. Mora-Medina et al. (2016) have recently reviewed sensory recognition (olfaction, vision, vocalization, hearing and direct contact) in relation to the ewe-lamb bond and emphasized the study demonstrated by Dwyer et al. (2003) that malnutrition of pregnant ewes during the gestation period impairs attachment between ewes and lambs by affecting maternal behaviors expressed at birth.

Data mining and its application in animal husbandry was studied by Wang et al. (2014). They underlined that animal husbandry management system structure is quite complex, with various problems faced by high volume and complex data; and some cannot establish a precise mathematical model.

For these problems, the application of data mining techniques can reflect a higher superiority, which is a powerful tool to solve such problems. Machine Learning (ML) and breeding share important objectives like prediction; and not surprisingly, several works have applied ML algorithms to genomic prediction (e.g., review Gonzalez-Recio et al., 2014). Sheep breeding use data sets and statistical techniques that qualify it for ML scope. In terms of lamb survival and factors affecting this very important parameter, ML methods can support us in creating predictive models by analyzing a large amount of data; and these methods can help us in decision-making. Machine learning researchers have developed sophisticated and effective algorithms which either complement or compete with the traditional statistical methods (Zupan et al., 2000).

The ability to study animal behaviour is important in many fields of science; and behavioral data represents large or open-ended data volumes which require machine learning techniques to automatically classify these large datasets into behavioural classes (Le Roux et al., 2017). In the scope of sheep breeding, ML algorithms have previously been used to detect basic behaviours (Fogarty et al., 2020a) such as mutually-exclusive behaviours (grazing, lying, standing, walking), active (or inactive) behaviour or detection of body posture (upright or prostrate). They reported that most effectively performing ML algorithms were Linear Kernel Support Vector Machine (SVM), Classification Tree (CART), and Linear Discriminant Analysis (LDA). Further studies have applied behaviour classification machine learning (ML) algorithms to accelerometer data to monitor changes in sheep behaviour around the time of lambing. It was aimed to facilitate the future development of algorithms based on ear tag accelerometer data for the detection of behavioural changes around the time of lambing in real-time or near-real-time (Fogarty et al., 2020b). More specific applications have analysed the relationships between serum lactoferrin concentrations and serum IgG concentrations in lambs (Gökçe et al., 2014) and automated detection of lameness in sheep using machine learning approaches (Kaler et al., 2020). These studies vary in their approach, they have differences in study purpose, maternal and offspring behavioral interactions, and dam and lamb intrinsic factors.

The study is divided into three parts. In the first one, we analyzed the behaviors of dam and lamb for target output of mothering ability for dam and successful sucking time for lambs. A comparison of classifiers was given in Table 2. In the second part of the study, we applied the PCA to reduce the dimensionality of the problem in order to check whether the results obtained with the whole set of variables are improved or not. The objective is to find the best classifier to predict lamb survival. In the last section, we tested the suitability of the different classifiers for lamb survival rate until weaning by using mothering ability and successful sucking time for lambs and excluded other mothering and lamb behavior data sets. The best two performers of the six classifiers that are applied for the prediction of mothering ability, lamb behavior and lamb survival at weaning is presented in Table 2. Results with whole sets of attributes were not improved with PCA; and classifiers were used without PCA reduction.

This paper provides an approach to machine learning algorithms in the behavioral (dam and offspring) and productive traits (dam age, dam live weight at lambing, dam breed, litter size, lamb birth weight and lamb sex) affecting lamb survival of native and crossbreed lambs produced in high altitudes and cold climate regions of Turkey. With this information, appropriate animal breeding and management programs can be formulated to reduce lamb mortality rates.