Dynamic forecasting of individual cow milk yield in automatic milking systems

Predicting age at first calving of dairy breed calves using whale optimization-based ensemble learning framework

Dairy product requirement and the demand-supply gap of milk in Ethiopia have been increasing at an alarming rate due to various factors such as shortage of animal's feeds, feed staffs, feed costs, and poor genetic merits of the local breeds of the country. This problem can be lessened by selecting best breed and modern animal breeding facilities, which require technologies like big data analysis and machine learning. In this study, a prediction model that can predict age at first calving of weaned calves based on their pre-weaning and weaning parameters, including dam's parity number, season of calving, birth weight, pre-weaning health status, pre-weaning average daily weight gain (ADG), weaning age and weaning weight is developed. Primary data collected by Ardayta Dairy Research Centre; Ethiopia is used for this research. First, different pre-trained models developed using support vector regression (SVR), Linear support vector regression (LSVR) and Nu support vector regression (NuSVR) techniques with their default hyperparameter values in which SVR performed best. Second, a model was developed by tuning hyperparameters of SVR including kernel function, regularization (C-parameter) and gamma parameters, and it resulted in an accuracy of 96.46%. Next, Whale optimization technique is used to select the optimized features of the dataset. Furthermore, an ensemble of SVR, LSVR, NuSVR is designed, and the framework is trained by optimized features of data. The designed model achieved an accuracy of 98.3% superseding the other combinations.

Predictive modelling of deviation from expected milk yield in transition cows on automatic milking systems

The transition period is a pivotal time in the production cycle of the dairy cow. It is estimated that between 30% and 50% of all cows experience metabolic or infectious disease during this time. One of the most common and economically consequential effects of disease during the transition period is a reduction in early lactation milk production. This has led to the utilisation of deviation from expected milk yield in early lactation as a proxy measure for transition health. However, to date, this analysis has been used exclusively for the retrospective assessment of transition cow health. Statistical models capable of predicting deviations from expected milk yield may allow producers to proactively manage animals predicted to suffer negative deviations in early lactation milk production. The objective of this retrospective cohort study was first, to explore the accuracy with which cow-level production and behaviour data collected on automatic milking systems (AMS) from 1-3 days in milk (DIM) can predict deviation from expected 30-day cumulative milk yield in multiparous cows. And second, to assess the accuracy with which predicted yield deviations can classify cows into groups which may facilitate improved transition management. Production, rumination, and physical activity data from 31 commercial AMS were accessed. A 3-step analytical procedure was then conducted. In Step 1, expected cumulative yield for 1-30 DIM for each individual cow-lactation was calculated using a mixed effect linear model. In Step 2, 30-Day Yield Deviation (YD) was calculated as the difference between observed and expected cumulative yield. Lactations were then assigned to one of three groups based on their YD, RED Group (0% YD). In Step 3, yield, rumination, and physical activity data from days 1-3 in lactation were used to predict YD using machine learning models. Following external validation, YD was predicted across the test data set with a mean absolute error of 9%. Categorisation of animals suffering large negative deviations (RED group) was achieved with a specificity of 99%, sensitivity of 35%, and balanced accuracy of 67%. Our results suggest that milk yield, rumination and physical activity patterns expressed by dairy cows from 1-3 DIM have utility in the prediction of deviation from expected 30-day cumulative yield. However, these predictions currently lack the sensitivity required to classify cows reliably and completely into groups which may facilitate improved transition cow management.

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.

The Potentialities of Machine Learning for Cow-Specific Milking: Automatically Setting Variables in Milking Machines

In dairy farming, milking-related operations are time-consuming and expensive, but are also directly linked to the farm’s economic profit. Therefore, reducing the duration of milking operations without harming the cows is paramount. This study aimed to test the variation in different parameters of milking operations on non-automatic milking machines to evaluate their effect on a herd and finally reduce the milking time. Two trials were set up on a dairy farm in Northern Italy to explore the influence of the pulsation ratio (60:40 vs. 65:35 pulsation ratio) and that of the detachment flow rate (600 g/min vs. 800 g/min) on milking performance, somatic cell counts, clinical mastitis, and teats score. Moreover, the innovative aspect of this study relates to the development of an optimized least-squares support vector machine (LSSVM) classification model based on the sparrow search algorithm (SSA) to predict the proper pulsation ratio and detachment flow rate for individual cows within the first two minutes of milking. The accuracy and precision of this model were 92% and 97% for shortening milking time at different pulsation ratios, and 78% and 79% for different detachment rates. The implementation of this algorithm in non-automatic milking machines could make milking operations cow-specific.

Short communication: The lag response of daily milk yield to heat stress in dairy cows

Previous studies suggest that there exists a lag relationship between daily milk yield and heat stress. The values of heat stress indicators (e.g., temperature-humidity index and ambient temperature) before test day have a simple correlation with daily milk yield on test day. However, the simple correlation might not be the best description because daily milk yield and heat stress indicators have a nature of time series in common, and their correlations are cross correlations that could be affected by autocorrelations. We hope to give a more reliable estimation on the lag relationship of daily milk yield via excluding autocorrelations with transfer function modeling. In this study, we found a lag relationship between daily milk yield and heat stress indicators based on transfer function modeling. Heat stress indicators included ambient temperature and temperature-humidity index. The daily milk yield data from 123 cows were obtained during a consecutive 63-d period (July 10-September 10, 2016). The mean daily milk yield (MY) and the maximum daily ambient temperature (TA_max) satisfied the stationary hypothesis, and the cross correlation between them was calculated. Before excluding autocorrelation, MY at 0 to 4 d after test day had significant cross correlations with TA_max on test day. After excluding the influence of autocorrelations, MY at 1 to 3 d after the test day had significant cross correlations with TA_max on test day. This result suggested that MY would respond to TA_max 1 d after the test day. In addition, the strength of cross correlations between MY and TA_max decreased from 1 to 3 d in sequence, implying a declining lag response of MY that would last for 3 d. The transfer function model for MY and TA_max is written as: MY_t = 16.90 + 0.74MY_{t- 1} - 0.25TA_max_{t- 1} + N_t, where N_t is white noise. This model can be used to track and predict the dynamic response of MY to TA_max.

Review: Application and Prospective Discussion of Machine Learning for the Management of Dairy Farms

Dairy farmers use herd management systems, behavioral sensors, feeding lists, breeding schedules, and health records to document herd characteristics. Consequently, large amounts of dairy data are becoming available. However, a lack of data integration makes it difficult for farmers to analyze the data on their dairy farm, which indicates that these data are currently not being used to their full potential. Hence, multiple issues in dairy farming such as low longevity, poor performance, and health issues remain. We aimed to evaluate whether machine learning (ML) methods can solve some of these existing issues in dairy farming. This review summarizes peer-reviewed ML papers published in the dairy sector between 2015 and 2020. Ultimately, 97 papers from the subdomains of management, physiology, reproduction, behavior analysis, and feeding were considered in this review. The results confirm that ML algorithms have become common tools in most areas of dairy research, particularly to predict data. Despite the quantity of research available, most tested algorithms have not performed sufficiently for a reliable implementation in practice. This may be due to poor training data. The availability of data resources from multiple farms covering longer periods would be useful to improve prediction accuracies. In conclusion, ML is a promising tool in dairy research, which could be used to develop and improve decision support for farmers. As the cow is a multifactorial system, ML algorithms could analyze integrated data sources that describe and ultimately allow managing cows according to all relevant influencing factors. However, both the integration of multiple data sources and the obtainability of public data currently remain challenging.

Performance of Online Somatic Cell Count Estimation in Automatic Milking Systems

Somatic cell count (SCC) is one of the most important and widely used mastitis diagnostics. For detecting (sub)clinical mastitis, online SCC related measurements are more and more used in automatic milking systems (AMS). Sensors such as an automated online California Mastitis Test (O-CMT) allow for high frequency screening of high SCC cows within a herd, which makes it potentially powerful to identify episodes of mastitis. However, the performance of O-CMT measurements, as compared to SCC determined in the laboratory (L-SCC), has only scarcely been described. The aims of this study were (1) to assess the agreement between the O-CMT measurement averaged over different time windows and the corresponding L-SCC measurements; (2) to determine the optimal time window for averaging O-CMT as compared to L-SCC; (3) to explore the added value of time-series of frequent O-CMT measurements in individual cow udder health monitoring compared to L-SCC measurements. Data were collected from 50 farms in 6 different countries that were equipped with AMS using O-CMT measurements and also performed regular L-SCC testing. We found that the overall concordance correlation coefficient (CCC) between O-CMT and L-SCC was 0.53 but differed substantially between farms. The CCC between O-CMT and L-SCC improved when averaging O-CMT over multiple milkings, with an optimal time-window of 24 h. Exploration of time series of daily O-CMT recordings show that this is an effective screening tool to find episodes of high SCC. Altogether, we conclude that although O-CMT agrees moderately with L-SCC, because of its high measurement frequency, it is a promising on-farm tool for udder health monitoring.

Unweaving tangled mortality and antibiotic consumption data to detect disease outbreaks - Peaks, growths, and foresight in swine production

As our capacity to collect and store health data is increasing, a new challenge of transforming data into meaningful information for disease monitoring and surveillance has arisen. The aim of this study was to explore the potential of using livestock mortality and antibiotic consumption data as a proxy for detecting disease outbreaks at herd level. Changes in the monthly records of mortality and antibiotic consumption were monitored in Danish swine herds that became positive for porcine reproductive and respiratory syndrome (PRRS) and porcine pleuropneumonia. Laboratory serological results were used to identify herds that changed from a negative to a positive status for the diseases. A dynamic linear model with a linear growth component was used to model the data. Alarms about state changes were raised based on forecast errors, changes in the growth component, and the values of the retrospectively smoothed values of the growth component. In all cases, the alarms were defined based on credible intervals and assessed prior and after herds got a positive disease status. The number of herds with alarms based on mortality increased by 3% in the 3 months prior to laboratory confirmation of PRRS-positive herds (Se = 0.47). A 22% rise in the number of weaner herds with alarms based on the consumption of antibiotics for respiratory diseases was found 1 month prior to these herds becoming PRRS-positive (Se = 0.22). For porcine pleuropneumonia-positive herds, a 10% increase in antibiotic consumption for respiratory diseases in sow herds was seen 1 month prior to a positive result (Se = 0.5). Monitoring changes in mortality data and antibiotic consumption showed changes at herd level prior to and in the same month as confirmation from diagnostic tests. These results also show a potential value for using these data streams as part of surveillance strategies.

Comparison of time-series models for monitoring temporal trends in endemic diseases sero-prevalence: lessons from porcine reproductive and respiratory syndrome in Danish swine herds

Background

Monitoring systems are essential to detect if the number of cases of a specific disease is rising. Data collected as part of voluntary disease monitoring programs is particularly useful to evaluate if control and eradication programs achieve the target. These data are characterized by random noise which makes harder to interpret temporal changes in the data. Monitoring trends in the data is a possible approach to overcome this issue.

The objective of this study was to assess the performance of three time-series models that allows monitoring trends in data in terms of its adaptability when used to monitor changes in disease sero-prevalence at a national scale based on data collected as part of voluntary monitoring programs. We compared two Bayesian forecasting methods and an Exponential smoothing method, specifically a Dynamic Linear Model, a Dynamic Generalized Linear Model and a Holt’s linear trend method, respectively. These three different types of time series models were applied to data on weekly sero-prevalence of Porcine Reproductive and Respiratory Syndrome (PRRS) in Danish swine herds.

Results

Comparing the linear cross-dependence between the filtered values obtained from the three models and the raw data, we observed that the Holt’s linear trend method shows negative linear dependence for roughly half of the time for breeding/nucleus and multiplier herds, having values close to zero for most of the period in finisher herds.

Conclusions

Bayesian forecasting methods adapt faster to changes in the data, compared to the deterministic Holt’s linear trend method. The practical implication of this greater flexibility is that the Bayesian methods will provide more reliable values of changes in the data and have potential to be implemented as part of a surveillance system in Denmark.

Electronic supplementary material

The online version of this article (10.1186/s12917-019-1981-y) contains supplementary material, which is available to authorized users.