Use of some data mining methods in akkaraman sheep breeding

In countries such as Turkey, where sheep breeding plays an important role, the conservation and breeding of indigenous sheep breeds is essential for preserving cultural heritage and supporting the rural economy. Early prediction of important traits such as growth, development and fertility in Akkaraman sheep and a better understanding of individual variations offer significant potential to breeders. This study analysed data collected from 36 farms engaged in Akkaraman sheep breeding, located in the Göçü and Yarma neighbourhoods of the Karatay district in Konya province. The dataset comprised birth, live weight, and survival records of 138,665 lambs born between 2012 and 2022, as well as live weight, reproductive performance (birth rate and type), and related yield records of 121,143 ewes from 2012 to 2021. This study investigated the potential of data mining algorithms to predict the individual yield performance of Akkaraman sheep. The research outlines the fundamental stages of the data mining process, details the criteria for algorithm parameter selection, and evaluates algorithm performance through applied examples based on actual phenotypic yield data. The least squares mean live weight of 104,498 Akkaraman ewes at the time of ram introduction was 53.16 kg, with year, age, and flock factors showing statistically significant effects (P<0.001). Similarly, the least squares mean birth weight of 108,015 Akkaraman lambs was 4.20 kg, and this trait was significantly influenced by year, sex, type of birth, dam age, and flock conditions (P<0.001). Furthermore, the least squares mean weaning weight at 120 days for 102,814 lambs was calculated as 31.66 kg, with year, sex, type of birth, dam age, and flock factors all having a statistically significant impacts on this trait (P<0.001). The analysis revealed that the average birth rate among 104,498 Akkaraman ewes was 90.87%, with flock, year, and age factors having statistically significant effects (P<0.05). The twin birth rate, an important indicator of fertility, was determined to be 13.83%, and this trait was significantly influenced by flock, year, and age of the ewe (P<0.05). Additionally, the survival rate of lambs until weaning was 95.18%, with flock, year, dam age, type of birth, and sex all showing significant effects (P<0.05). Among the evaluated factors, flock exhibited the greatest variation with phenotypic yield traits. In this study, tree-based algorithms such as AdaBoost, XGBoost, Random Forest, Regression Tree, CHAID, and Exhaustive CHAID, as well as distance-based algorithms including Artificial Neural Networks (ANN) and k-Nearest Neighbours (k-NN), were employed to predict continuous dependent variables derived from performance records (ewe live weight, lamb birth weight, and lamb weaning weight). The algorithms did not demonstrate high overall predictive performance based on goodness of fit criteria; however, the best results were obtained for the prediction of weaning weight. The findings suggest that the XGBoost and ANN algorithms have potential for effectively predicting weaning live weight in Akkaraman lambs. AdaBoost, Random Forest, k-NN, Naive Bayes, Logistic Regression, ANN, KStar, Ridor, Classification Tree (J48), CHAID and Exhaustive CHAID algorithms were used for the prediction of binary categorical yield traits (Lambed/Not Lambed, Single/Twin and Dead/Live). In the study using balanced data set, it was determined that the algorithms predicted the infertile and twin births of ewes with an accuracy rate of 70%. The probability of lambs dying was predicted by the algorithms with an accuracy rate of 60%. It is thought that Logistic Regression, Naive Bayes and Random Forest algorithms can be used in similar studies.