Forecasting Seasonal Milk Production Using MARS Algorithm for Multiple Continuous Responses in Holstein Dairy Cattles

Abstract

In this study, seasonal milk yield estimation will be made using multivariate adaptive regression spline (MARS) algorithm for multiple continuous responses in dairy cattle (Holstein hybrid). For the research, milking records for the years 2020-2021 were collected from 157 dairy animals using Holstein hybrid dairy cattle from a research farm in Konya, Türkiye. The amount of feed given in this experiment was not changed and the effect of the season on the estimation of milk yield was investigated in the study. The analyzed independent variables used in the study were pregnancy status (PS), number of days milked (MDN), Lactation Number (LN), age of cows (months), average seven-day milk yield (7-Day Average Milk-SDMY), last lactation milk yield (last_MY), number of inseminations (IN), peak yield (Pik_Yield) and target variables were calculated as (YieldAutumn/winter/spring/summer (kg) = Mean milk mean of season. In this context, the ehaGoF package was used to measure the prediction performance of the simultaneous MARS model established with the earth package for MARS analysis. MARS estimation equations obtained simultaneously for four dependent variables (multiple responses) are given. By looking at the MARS equation, the MARS model estimation equation was determined for the optimum milk yield, the threshold values, the three threshold values determined in the model were determined as MDN, Age, Peak_Yield, and the corresponding values were respectively; 159 days, 39.6 (months) and 37.1 kg/day. Considering the estimation equation, it is seen that the independent variables MDN, SDMY and LN are the most important variables in determining the estimation equation. It is seen that the best fitting value for the estimation equation of the dependent variables is the YieldWinter variable.