SmartBreed: Smart animal breeding with advanced machine learning

Accurate prediction of animals’ future phenotype, especially at an early age, can be beneficial to management decisions and interventions. The phenotypic information relevant for future prediction is accumulating throughout the animals’ productive life and technology (e.g. sensors) to record this information is emerging. At an early age phenotypic information might however be limited, but with the advent of genomics, accurate genetic information is available to possibly fill this gap. In this project we investigate possibilities to predict future performance based on genomic and phenotypic information using state-of-the-art methods of prediction.  

In this study we predicted slaughter weight in pigs at start of the finishing phase based on phenotypic and genomic features. Machine learning methods allowed the integration of phenotypic and genetic features in a single predictive model that performed better than models built with either phenotypic or genetic features alone. The developed method outperformed standard pig assignment strategies and can assist farmers in creating uniform pens, achieving a classification error of 0.33 in k-fold cross validation compared to 0.56 for the traditional methods. In this application pedigree and genomic pairwise distances as well as EBVs could be used interchangeably in the predictions.

Machine learning was used to classify pigs into one of 5 classes of muscularity based solely on depth images obtained by a Kinect sensor while the pig freely passed through a corridor. A protocol was developed comprising automized picture selection, filtering and prediction. Compared to subjective muscle scoring the mean absolute error rate amounted to 0.65.    

A general problem in machine learning applications is that performance is suboptimal when redundancy exists among features. In this study we developed an alternative method of feature selection and proofed that prediction was improved after application of this method.

In dairy cattle we investigated the possibilities of predicting survival to 2nd lactation at different moments in the productive life of a cow. We combined genomic breeding values with the accumulating phenotypic data and prediction was performed using alternative machine learning techniques as well as logistic regression. The predictive power was moderate and neither of the methods applied had best performance at all times. In all cases however a combination of genomic breeding values and phenotypic information gave the best predictions, esp. at young ages genomic information had a considerable contribution. 

We subsequentially investigated if we could improve prediction of cow survival to second lactation by combining the predictions of multiple (weak) methods in an ensemble method. We tested four ensemble methods which were evaluated on five different performance metrics. From the ensemble methods being tested the multiple logistic regression ensemble method performed best on most of the performance metrics. 

Published results in this project are:

van der Heide, E. M. M., R. F. Veerkamp, M. L. van Pelt, C. Kamphuis, and B. J. Ducro. 2020. Predicting survival in dairy cattle by combining genomic breeding values and phenotypic information. J. Dairy Sci. 103:556-571.

van der Heide, E. M. M., R. F. Veerkamp, M. L. van Pelt, C. Kamphuis, I. Athanasiadis, and B. J. Ducro. 2019. Comparing regression, naive Bayes, and random forest methods in the prediction of individual survival to second lactation in Holstein cattle. J. Dairy Sci. 102:9409-9421.

Alsahaf, A., G. Azzopardi, B. Ducro, E. Hanenberg, R. F. Veerkamp, and N. Petkov. 2019. Estimation of muscle scores of live pigs using a Kinect camera. IEEE Access. 7:52238-52245.

Alsahaf, A., G. Azzopardi, B. Ducro, E. Hanenberg, R. F. Veerkamp, and N. Petkov. 2018. Prediction of slaughter age in pigs and assessment of the predictive value of phenotypic and genetic information using random forest. J. Anim. Sci. 96:4935-4943.


Project leader: This email address is being protected from spambots. You need JavaScript enabled to view it.

Project supporters:

project partner Cobb  project partner CRV  project partner Hendrix Genetics  project partner Topigs Norsvin

Check out our Partners