Predicción de la producción diaria de leche enbovinos Gyr a través de métodos de aprendizaje supervisado

The Asociaci ́on Colombiana de Criadores de Ganado Ceb ́u - ASOCEBU, has in-terest in developing a machine to predict total daily milk yield using partial pro-duction measurements in Gyr cattle and, in particular, answering two questions: 1)can a reference predictive method be outperformed by locally developed methods?2) which one of the two partial records (AM or PM) has a better predictive perfor-mance? Therefore, the objective of this paper was to develop a predictive machinefor daily milk yield in Gyr cattle using partial records, milking interval, days inmilk, and parity (n=13806), by implementing supervised learning methods. Besi-des the reference predictive machine, several combinations of input variables andmodel or learning method were considered. Artificial neural networks, supportvector machines, random forests, and linear regression with location parametersestimated via least squares, or the shrinkage methods Ridge and Lasso were used.The predictive performance (PP) was assessed through crossvalidation using thefollowing error functions: square root of mean square error (RMSE) and meanabsolute error (MAE). It was found that an artificial neural network with a singlehidden layer and the AM partial record, milking interval, parity and days in milkas input variables had the best PP (RMSE=1.5042, MAE=1.1389), but in gene-ral, the performance of the methods was similar. All machines whose parameterswere learned using local data outperformed the reference method and the morningpartial records showed a better PP than those from the afternoon. These resultspermit guiding ASOCEBU’s milk control program and generate a “tailormade”method to predict total daily milk yield of Gyr cattle in Colombia, a relevantcomponent of the genetic improvement and productivity modelling programs ofthis breed. La Asociaci ́on Colombiana de Criadores de Ganado Ceb ́u - ASOCEBU, tiene in-ter ́es en desarrollar una m ́aquina para predecir la producci ́on total diaria de lecheempleando mediciones de producci ́on parciales en ganado Gyr y, en particular,responder dos preguntas: 1) ¿puede un m ́etodo predictivo de referencia ser supera-do por m ́etodos desarrollados a nivel local? 2) ¿cu ́al de los dos registros parciales(AM o PM) tiene un mejor desempe ̃no predictivo? Por lo tanto, el objetivo deeste art ́ıculo fue desarrollar una m ́aquina predictiva para la producci ́on diaria deleche en bovinos Gyr utilizando registros parciales, intervalo entre orde ̃nos, d ́ıas enlactancia