Forecasting the milk yield of cows on farms equipped with automatic milking system with the use of decision trees

Comparison of selected parameters of automated milking in dairy cattle barns equipped with a concentrate feeding system

Automatic milking systems (AMSs) give cows relative freedom to choose the time and frequency of milking throughout the day. Feeding stations also may improve the management of farms. Combining milking robots and feeding stations (FS) may improve milking efficiency and milk yield. Therefore, combining AMS and FS may be beneficial for farmers. The objective of the research was to compare selected automatic milking parameters (daily indices per cow) registered by an AMS in relation to selected features including the presence of concentrate feeding stations. We analysed 931 cows born in 2013-14, in lactations 1-8. In total, we collected data from 357 318 milking days. The following parameters were examined: milking frequency (n/24 h), number of rejected milking (n/24 h), the average number of nipple attempts (n/milking), milking speed (kg/min), time spent in the milking box (s/24 h), milk yield (kg/24 h), milking efficiency (kg/min), rumination time (min/24 h), and concentrate intake (kg) per 100 kg of milk produced. The statistical analysis was conducted using a multi-factor analysis of variance. The analysis confirmed a statistical effect of the concentrate feeding system on most of the investigated traits, except for nipple attempts, box time and rumination time. In cows in barns with an FS, the following parameters were statistically higher compared to cows in non-FS barns: milking frequency (3.04 vs 2.73n/24 h), number of rejected milking (2.24 vs 1.51n/24 h), milking speed (2.98 vs 2.64 kg/min), milk yield (33.48 vs 30.14 kg/24 h), milking efficiency (1.80 vs 1.67 kg/min), and concentrate intake per 100 kg of milk produced (14.67 vs 12.67 kg). The study results indicate that using feeding stations in combination with an AMS can increase milking efficiency, hence the milk output from a milking robot.

A Literature Review of Modeling Approaches Applied to Data Collected in Automatic Milking Systems

Automatic milking systems (AMS) have played a pioneering role in the advancement of Precision Livestock Farming, revolutionizing the dairy farming industry on a global scale. This review specifically targets papers that focus on the use of modeling approaches within the context of AMS. We conducted a thorough review of 60 articles that specifically address the topics of cows' health, production, and behavior/management Machine Learning (ML) emerged as the most widely used method, being present in 63% of the studies, followed by statistical analysis (14%), fuzzy algorithms (9%), deterministic models (7%), and detection algorithms (7%). A significant majority of the reviewed studies (82%) primarily focused on the detection of cows' health, with a specific emphasis on mastitis, while only 11% evaluated milk production. Accurate forecasting of dairy cow milk yield and understanding the deviation between expected and observed milk yields of individual cows can offer significant benefits in dairy cow management. Likewise, the study of cows' behavior and herd management in AMSs is under-explored (7%). Despite the growing utilization of machine learning (ML) techniques in the field of dairy cow management, there remains a lack of a robust methodology for their application. Specifically, we found a substantial disparity in adequately balancing the positive and negative classes within health prediction models.

Genomic-based genetic parameters for milkability traits derived from automatic milking systems in North American Holstein cattle

The number of dairy farms adopting automatic milking systems (AMS) has considerably increased around the world aiming to reduce labor costs, improve cow welfare, increase overall performance, and generate a large amount of daily data, including production, behavior, health, and milk quality records. In this context, this study aimed to (1) estimate genomic-based variance components for milkability traits derived from AMS in North American Holstein cattle based on random regression models; and (2) derive and estimate genetic parameters for novel behavioral indicators based on AMS-derived data. A total of 1,752,713 daily records collected using 36 milking robot stations and 70,958 test-day records from 4,118 genotyped Holstein cows were used in this study. A total of 57,600 SNP remained after quality control. The daily-measured traits evaluated were milk yield (MY, kg), somatic cell score (SCS, score unit), milk electrical conductivity (EC, mS), milking efficiency (ME, kg/min), average milk flow rate (FR, kg/min), maximum milk flow rate (FRM, kg/min), milking time (MT, min), milking failures (MFAIL), and milking refusals (MREF). Variance components and genetic parameters for MY, SCS, ME, FR, FRM, MT, and EC were estimated using the AIREMLF90 software under a random regression model fitting a third-order Legendre orthogonal polynomial. A threshold Bayesian model using the THRGIBBS1F90 software was used for genetically evaluating MFAIL and MREF. The daily heritability estimates across days in milk (DIM) ranged from 0.07 to 0.28 for MY, 0.02 to 0.08 for SCS, 0.38 to 0.49 for EC, 0.45 to 0.56 for ME, 0.43 to 0.52 for FR, 0.47 to 0.58 for FRM, and 0.22 to 0.28 for MT. The estimates of heritability (± SD) for MFAIL and MREF were 0.02 ± 0.01 and 0.09 ± 0.01, respectively. Slight differences in the genetic correlations were observed across DIM for each trait. Strong and positive genetic correlations were observed among ME, FR, and FRM, with estimates ranging from 0.94 to 0.99. Also, moderate to high and negative genetic correlations (ranging from -0.48 to -0.86) were observed between MT and other traits such as SCS, ME, FR, and FRM. The genetic correlation (± SD) between MFAIL and MREF was 0.25 ± 0.02, indicating that both traits are influenced by different sets of genes. High and negative genetic correlations were observed between MFAIL and FR (-0.58 ± 0.02) and MFAIL and FRM (-0.56 ± 0.02), indicating that cows with more MFAIL are those with lower FR. The use of random regression models is a useful alternative for genetically evaluating AMS-derived traits measured throughout the lactation. All the milkability traits evaluated in this study are heritable and have demonstrated selective potential, suggesting that their use in dairy cattle breeding programs can improve dairy production efficiency in AMS.

Forecasting Milking Efficiency of Dairy Cows Milked in an Automatic Milking System Using the Decision Tree Technique

In barns equipped with an automatic milking system, the profitability of production depends primarily on the milking efficiency of a cow (ME; kg/min) defined as cow milk yield per minute of box time. This study was carried out on 1823 Polish Holstein−Friesian cows milked by the automatic milking system (AMS) in 20 herds. Selected milking parameters recorded by the AMS were analyzed in the research. The aim of the study was to forecast ME using two statistical techniques (analysis of variance and decision trees). The results of the analysis of variance showed that the average ME was 1.67 kg/min. ME was associated with: year of AMS operation (being the highest in the first year), number of cows per robot (the highest in robots with 61−75 cows), lactation number (highest for multiparas), season of calving (the highest in spring), age at first calving (>36 months), days in milk (151−250 days) and finally, rear quarter to total milk yield ratio (the highest between 51% and 55%). The decision tree predicted that the highest ME (2.01 kg/min) corresponded with cows that produced more than 45 kg of milk per day, were milked less than four times/day, had a short teatcup attachment time (<7.65 s) and were milked in robots that had an occupancy lower than 56 cows.

Implementation of Technical and Technological Progress in Dairy Production

The involvement of people and technical devices is a characteristic feature of technological processes in agriculture. Human access to modernized and more efficient technical equipment determines the differentiation of the proportions of the contributions of human labor and technical equipment to the implementation of production technology on farms. Taking into account the data on manual and machine work inputs, the methodology of determining the technological index level (TL) was presented. The aim of the present study was to present the scope of use of the technological index level to assess the effects of technological progress in the dairy production system, with particular emphasis on cow milking. For the value range of the technological index level (0–100%), changes in the milkman’s work efficiency were presented based on research carried out on farms equipped with milking equipment at different levels of technical advancement. Moreover, the course of changes in electricity and water consumption per liter of milk was determined in association with the technological index level. The issue of simultaneous implementation of various forms of progress was developed based on the example of milking cows with a milking robot. Five categories (ranges) of cows’ milk yield were distinguished and compared with the current yields of cows in the European Union. On this basis, a discussion was initiated on the factors that facilitate and limit the implementation of technical and technological progress in dairy production.

Prediction of cow's fertility based on data recorded by automatic milking system during the periparturient period

The results of most studies show the beneficial effect of milking automation on production parameters of dairy cows, but its effect on fertility traits is debatable. Therefore, a study was undertaken to predict cow fertility - services per conception (SC) and calving interval (CI) - based on automatic milking system (AMS) data collected in the periparturient period subdivided into the second and first week before calving, 1-4, 5-7, 8-14, 15-21 and 22-28 days of lactation. SC and CI were predicted using daily indicators such as concentrate intake, number of milkings, cow box time, milking time, milking speed, colostrum and milk yield, composition, temperature and electrical conductivity. The study material was derived from the AMS management system and from the SYMLEK milk recording system. The analysis covered data for 16,329 milkings of 398 Polish Holstein-Friesian (PHF) cows, which were used in three AMS herds. The collected numerical data were statistically analysed by correlation analysis in parallel with decision tree technique (SAS statistical package). The present study showed that due to the low, mostly non-significant coefficients of correlation between AMS data collected between 2 weeks before and 4 weeks after calving, it is not possible to predict cow fertility based on single traits. It has been established that the decision tree method may help breeders, already during the postcalving period, to choose the level of factors associated with AMS milking, which will ensure good fertility of cows in a herd. The most favourable number of services per conception is to be expected from cows that were milked <1.6 times per day from 1 to 4 days of lactation and electrical conductivity of their colostrum did not exceed 69 mS during that time. In turn, shortest CI (366 days) will be characteristic of the cows whose average daily colostrum yield did not exceed 20.2 kg and their daily concentrate intake from 8 to 14 days of lactation was at least 5.0 kg.

Random Forest Modelling of Milk Yield of Dairy Cows under Heat Stress Conditions

Precision Livestock Farming (PLF) relies on several technological approaches to acquire, in the most efficient way, precise and real-time data concerning production and welfare of individual animals. In this regard, in the dairy sector, PLF devices are being increasingly adopted, automatic milking systems (AMSs) are becoming increasingly widespread, and monitoring systems for animals and environmental conditions are becoming common tools in herd management. As a consequence, a great amount of daily recorded data concerning individual animals are available for the farmers and they could be used effectively for the calibration of numerical models to be used for the prediction of future animal production trends. On the other hand, the machine learning approaches in PLF are nowadays considered an extremely promising solution in the research field of livestock farms and the application of these techniques in the dairy cattle farming would increase sustainability and efficiency of the sector. The study aims to define, train, and test a model developed through machine learning techniques, adopting a Random Forest algorithm, having the main goal to assess the trend in daily milk yield of a single cow in relation to environmental conditions. The model has been calibrated and tested on the data collected on 91 lactating cows of a dairy farm, located in northern Italy, and equipped with an AMS and thermo-hygrometric sensors during the years 2016-2017. In the statistical model, having seven predictor features, the daily milk yield is evaluated as a function of the position of the day in the lactation curve and the indoor barn conditions expressed in terms of daily average of the temperature-humidity index (THI) in the same day and its value in each of the five previous days. In this way, extreme hot conditions inducing heat stress effects can be considered in the yield predictions by the model. The average relative prediction error of the milk yield of each cow is about 18% of daily production, and only 2% of the total milk production.

Prediction of Lactational Milk Yield of Cows Based on Data Recorded by AMS during the Periparturient Period

Simple Summary

Barns equipped with the automatic milking system (AMS) record huge amounts of data on milk flow rate, milk yield and composition, milk temperature, amount of concentrate intake and rumination time. Our study attempted to use this information, recorded during the periparturient period (divided into subperiods: second (14–8 days) and first (7–1 days) week before calving; 1–4, 5–7, 8–14, 15–21 and 22–28 days of lactation), to predict lactation milk yield in Polish Holstein–Friesian cows. In the first stage of statistical analysis, coefficients of simple correlation between lactation milk yield and AMS parameters were calculated. We found that prediction of lactation milk yield based on individual pieces of data may be ineffective—the calculated coefficients of correlation were low or moderate. In the next step of data analysis, we used a modern data mining technique in the form of decision trees. Based on the graphic, easy-to-interpret decision tree, we concluded that the highest lactation yield is to be expected for cows with completed lactations (survived until the next lactation), which were milked 4.07 times per day on average in the 4th week of lactation.

Abstract

Early prediction of lactation milk yield enables more efficient herd management. Therefore, this study attempted to predict lactation milk yield (LMY) in 524 Polish Holstein–Friesian cows, based on information recorded by the automatic milking system (AMS) in the periparturient period. The cows calved in 2016 and/or 2017 and were used in 3 herds equipped with milking robots. In the first stage of data analysis, calculations were made of the coefficients of simple correlation between rumination time (expressed as mean time per cow during the periparturient period: second (14–8 days) and first (7–1 days) week before calving, 1–4, 5–7, 8–14, 15–21 and 22–28 days of lactation), electrical conductivity and temperature of milk (expressed as means per cow on days 1–4, 5–7, 8–14, 15–21 and 22–28), amount of concentrate intake, number of milkings/day, milking time/visit, milk speed and lactation milk yield. In the next step of the statistical analysis, a decision tree technique was employed to determine factors responsible for LMY. The study showed that the correlation coefficients between LMY and AMS traits recorded during the periparturient period were low or moderate, ranging from 0.002 to 0.312. Prediction of LMY from the constructed decision tree model was found to be possible. The employed Classification and Regression Trees (CART) algorithm demonstrated that the highest lactation yield is to be expected for cows with completed lactations (survived until the next lactation), which were milked 4.07 times per day on average in the 4th week of lactation. We proved that the application of the decision tree method could allow breeders to select, already in the postparturient period, appropriate levels of AMS milking variables, which will ensure high milk yield per lactation.