Machine Learning to Predict Pregnancy in Dairy Cows: An Approach Integrating Automated Activity Monitoring and On-Farm Data

Automated activity monitoring (AAM) systems are critical in the dairy industry for detecting estrus and optimizing the timing of artificial insemination (AI), thus enhancing pregnancy success rates in cows. This study developed a predictive model to improve pregnancy success by integrating AAM data with cow-specific and environmental factors. Utilizing data from 1,054 cows, this study compared the pregnancy outcomes between two AI timings-8 or 10 h post-AAM alarm. Variables such as age, parity, body condition, locomotion, and vaginal discharge scores, peripartum diseases, the breeding program, the bull used for AI, milk production at the time of AI, and environmental conditions (season, relative humidity, and temperature-humidity index) were considered alongside the AAM data on rumination, activity, and estrus intensity. Six predictive models were assessed to determine their efficacy in predicting pregnancy success: logistic regression, Bagged AdaBoost algorithm, linear discriminant, random forest, support vector machine, and Bagged Classification Tree. Integrating the on-farm data with AAM significantly enhanced the pregnancy prediction accuracy at AI compared to using AAM data alone. The random forest models showed a superior performance, with the highest Kappa statistic and lowest false positive rates. The linear discriminant and logistic regression models demonstrated the best accuracy, minimal false negatives, and the highest area under the curve. These findings suggest that combining on-farm and AAM data can significantly improve reproductive management in the dairy industry.

Prediction of persistency for day 305 of lactation at the moment of the insemination decision

When deciding on the voluntary waiting period of an individual cow, it might be useful to have insight into the persistency for the remainder of that lactation at the moment of the insemination decision, especially for farmers who consider persistency in their reproduction management. Currently, breeding values for persistency are calculated for dairy cows but, to our knowledge, prediction models to accurately predict persistency at different moments of insemination are lacking. This study aimed to predict lactation persistency for DIM 305 at different insemination moments (DIM 50, 75, 100, and 125). Available cow and herd level data from 2005 to 2022 were collected for a total of 20,508 cows from 85 herds located in the Netherlands and Belgium. Lactation curve characteristics were estimated for every daily record using the data up to and including that day. Persistency was defined as the number of days it takes for the milk production to decrease by half during the declining stage of lactation, and calculated from the estimated lactation curve characteristic 'decay'. Four linear regression models for each of the selected insemination moment were built separately to predict decay at DIM 305 (decay-305). Independent variables included the lactation curve characteristics at the selected insemination moment, daily milk yield, age, calving season, parity group and other herd variables. The average decay-305 of primiparous cows was lower than that of multiparous cows (1.55 *10-3 vs. 2.41*10-3, equivalent to a persistency of 447 vs. 288 days, respectively). Results showed that our models had limitations in accurately predicting persistency, although predictions improved slightly at later insemination moments, with R2 values ranging between 0.27 and 0.41. It can thus be concluded that, based only on cow and herd milk production information, accurate prediction of persistency for DIM 305 is not feasible.

Cow Behavioural Activities in Extensive Farms: Challenges of Adopting Automatic Monitoring Systems

Animal welfare is becoming an increasingly important requirement in the livestock sector to improve, and therefore raise, the quality and healthiness of food production. By monitoring the behaviour of the animals, such as feeding, rumination, walking, and lying, it is possible to understand their physical and psychological status. Precision Livestock Farming (PLF) tools offer a good solution to assist the farmer in managing the herd, overcoming the limits of human control, and to react early in the case of animal health issues. The purpose of this review is to highlight a key concern that occurs in the design and validation of IoT-based systems created for monitoring grazing cows in extensive agricultural systems, since they have many more, and more complicated, problems than indoor farms. In this context, the most common concerns are related to the battery life of the devices, the sampling frequency to be used for data collection, the need for adequate service connection coverage and transmission range, the computational site, and the performance of the algorithm embedded in IoT-systems in terms of computational cost.

A Recognition Method of Ewe Estrus Crawling Behavior Based on Multi-Target Detection Layer Neural Network

There are some problems with estrus detection in ewes in large-scale meat sheep farming: mainly, the manual detection method is labor-intensive and the contact sensor detection method causes stress reactions in ewes. To solve the abovementioned problems, we proposed a multi-objective detection layer neural network-based method for ewe estrus crawling behavior recognition. The approach we proposed has four main parts. Firstly, to address the problem of mismatch between our constructed ewe estrus dataset and the YOLO v3 anchor box size, we propose to obtain a new anchor box size by clustering the ewe estrus dataset using the K-means++ algorithm. Secondly, to address the problem of low model recognition precision caused by small imaging of distant ewes in the dataset, we added a 104 × 104 target detection layer, making the total target detection layer reach four layers, strengthening the model's ability to learn shallow information and improving the model's ability to detect small targets. Then, we added residual units to the residual structure of the model, so that the deep feature information of the model is not easily lost and further fused with the shallow feature information to speed up the training of the model. Finally, we maintain the aspect ratio of the images in the data-loading module of the model to reduce the distortion of the image information and increase the precision of the model. The experimental results show that our proposed model has 98.56% recognition precision, while recall was 98.04%, F1 value was 98%, mAP was 99.78%, FPS was 41 f/s, and model size was 276 M, which can meet the accurate and real-time recognition of ewe estrus behavior in large-scale meat sheep farming.

The Marginal Abatement Cost of Antimicrobials for Dairy Cow Mastitis: A Bioeconomic Optimization Perspective

Maintaining udder health is the primary indication for antimicrobial use (AMU) in dairy production, and modulating this application is a key factor in decreasing AMU. Defining the optimal AMU and the associated practical rules is challenging since AMU interacts with many parameters. To define the trade-offs between decreased AMU, labor and economic performance, the bioeconomic stochastic simulation model DairyHealthSim (DHS)© was applied to dairy cow mastitis management and coupled to a mean variance optimization model and marginal abatement cost curve (MACC) analysis. The scenarios included three antimicrobial (AM) treatment strategies at dry-off, five types of general barn hygiene practices, five milking practices focused on parlor hygiene levels and three milk withdrawal strategies. The first part of economic results showed similar economic performances for the blanked dry-off strategy and selective strategy but demonstrated the trade-off between AMU reduction and farmers' workload. The second part of the results demonstrated the optimal value of the animal level of exposure to AM (ALEA). The MACC analysis showed that reducing ALEA below 1.5 was associated with a EUR 10,000 loss per unit of ALEA on average for the farmer. The results call for more integrative farm decision processes and bioeconomic reasoning to prompt efficient public interventions.

Symposium review: Use of multiple biological, management, and performance data for the design of targeted reproductive management strategies for dairy cows

As the reproductive efficiency of dairy cattle continues to improve in response to better management and use of technology, novel reproductive management approaches will be required to improve herd performance, profitability, and sustainability. A potential approach currently being explored is targeted reproductive management. This approach consists of identifying cows with different reproductive and performance potential using multiple traditional and novel sources of biological, management, and performance data. Once subgroups of cows that share biological and performance features are identified, reproductive management strategies specifically designed to optimize cow performance, herd profitability, or alternative outcomes of interest are implemented on different subgroups of cows. Tailoring reproductive management to subgroups of cows is expected to generate greater gains in outcomes of interest than if the whole herd is under similar management. Major steps in the development and implementation of targeted reproductive management programs for dairy cattle include identification and validation of robust predictors of reproductive outcomes and cow performance, and the development and on-farm evaluation of reproductive management strategies for optimizing outcomes of interest for subgroups of cows. Predictors of cow performance currently explored for use in targeted management include genomic predictions; behavioral, physiological, and performance parameters monitored by sensor technologies; and individual cow and herd performance records. Once the most valuable predictive sources of variation are identified and their effects quantified, novel analytic methods (e.g., machine learning) for prediction will likely be required. These tools must identify groups of cows for targeted management in real time and with no human input. Despite some encouraging research evidence supporting the development of targeted reproductive management strategies, extensive work is required before widespread implementation by commercial farms.

A comparison of machine learning and logistic regression in modelling the association of body condition score and submission rate

The effect of body condition score (BCS) on reproductive outcomes is complex, dynamic and non-linear with interaction and confounding. The flexibility inherent in machine learning algorithms makes them attractive for analysing complex data. This study was designed to compare the ability of a range of machine learning techniques in estimating the probability of service within 21 days of the planned start of mating. We hypothesised that if there were complex and unknown interactions or non-linearity in the data, some machine learning algorithms would result in superior model performance compared to regression models. For a period of six months from the planned start of calving, BCS was visually assessed once a month for 6127 cows on 8 commercial New Zealand dairy farms by a trained veterinarian using the DairyNZ 10-point range for every cow in the herd. Cow, lactation and reproductive data was extracted from the national herd database. This data was used to predict probability of service within 21 days of planned start of mating (PSM) using mixed multivariable logistic regression and decision tree, k-nearest neighbour, random forest and neural network analysis. Models were adjusted for herd, cow age, breed, days in milk, BCS at calving, BCS change between calving and mating, BCS change after mating, volume adjusted milk protein and fat concentration pre-mating. Models were constructed on a training data set using 10-fold cross validation repeated 10 times and evaluated on a test data set using discrimination and calibration techniques. In all models, days calved at PSM was the most important variable for predicting submission rate, followed by BCS at PSM. Factors associated with an increased probability of insemination were calving at a BCS of 5.0, losing less BCS after calving, having a higher BCS at nadir, losing BCS rapidly after calving, nadir occurring before PSM and calving early. All the models except for the decision tree had an area under the receiver operating characteristic curve (AUC) in the range 0.68-0.73 indicating good overall discriminatory power, but calibration analysis suggested all models were better at predicting cows that got inseminated than correctly identifying animals that did not get inseminated. Overall, the machine learning techniques were no better than a generalised logistic regression model. These results highlight the importance of BCS targets at calving and indicate BCS loss, milk characteristics and days calves may be useful indicators identifying cows at risk of poor reproductive outcomes.

Uncovering sperm metabolome to discover biomarkers for bull fertility

Background

Subfertility decreases the efficiency of the cattle industry because artificial insemination employs spermatozoa from a single bull to inseminate thousands of cows. Variation in bull fertility has been demonstrated even among those animals exhibiting normal sperm numbers, motility, and morphology. Despite advances in research, molecular and cellular mechanisms underlying the causes of low fertility in some bulls have not been fully elucidated. In this study, we investigated the metabolic profile of bull spermatozoa using non-targeted metabolomics. Statistical analysis and bioinformatic tools were employed to evaluate the metabolic profiles high and low fertility groups. Metabolic pathways associated with the sperm metabolome were also reported.

Results

A total of 22 distinct metabolites were detected in spermatozoa from bulls with high fertility (HF) or low fertility (LF) phenotype. The major metabolite classes of bovine sperm were organic acids/derivatives and fatty acids/conjugates. We demonstrated that the abundance ratios of five sperm metabolites were statistically different between HF and LF groups including gamma-aminobutyric acid (GABA), carbamate, benzoic acid, lactic acid, and palmitic acid. Metabolites with different abundances in HF and LF bulls had also VIP scores of greater than 1.5 and AUC- ROC curves of more than 80%. In addition, four metabolic pathways associated with differential metabolites namely alanine, aspartate and glutamate metabolism, β-alanine metabolism, glycolysis or gluconeogenesis, and pyruvate metabolism were also explored.

Conclusions

This is the first study aimed at ascertaining the metabolome of spermatozoa from bulls with different fertility phenotype using gas chromatography-mass spectrometry. We identified five metabolites in the two groups of sires and such molecules can be used, in the future, as key indicators of bull fertility.

Comparing regression, naive Bayes, and random forest methods in the prediction of individual survival to second lactation in Holstein cattle

In this study, we compared multiple logistic regression, a linear method, to naive Bayes and random forest, 2 nonlinear machine-learning methods. We used all 3 methods to predict individual survival to second lactation in dairy heifers. The data set used for prediction contained 6,847 heifers born between January 2012 and June 2013, and had known survival outcomes. Each animal had 50 genomic estimated breeding values available at birth and up to 65 phenotypic variables that accumulated over time. Survival was predicted at 5 moments in life: at birth, at 18 mo, at first calving, at 6 wk after first calving, and at 200 d after first calving. The data sets were randomly split into 70% training and 30% testing sets to evaluate model performance for 20-fold validation. The methods were compared for accuracy, sensitivity, specificity, area under the curve (AUC) value, contrasts between groups for the prediction outcomes, and increase in surviving animals in a practical scenario. At birth and 18 mo, all methods had overlapping performance; no method significantly outperformed the other. At first calving, 6 wk after first calving, and 200 d after first calving, random forest and naive Bayes had overlapping performance, and both machine-learning methods outperformed multiple logistic regression. Overall, naive Bayes has the highest average AUC at all decision points up to 200 d after first calving. Random forest had the highest AUC at 200 d after first calving. All methods obtained similar increases in survival in the practical scenario. Despite this, the methods appeared to predict the survival of individual heifers differently. All methods improved over time, but the changes in mean model outcomes for surviving and non-surviving animals differed by method. Furthermore, the correlations of individual predictions between methods ranged from r = 0.417 to r = 0.700; the lowest correlations were at first calving for all methods. In short, all 3 methods were able to predict survival at a population level, because all methods improved survival in a practical scenario. However, depending on the method used, predictions for individual animals were quite different between methods.

Hyperketolactia occurrence before or after artificial insemination is associated with a decreased pregnancy per artificial insemination in dairy cows

The reproductive parameters of dairy cattle have continuously declined worldwide over the last 50 years. Nutritional imbalances are identified as risk factors for this decrease of reproductive performance. The present paper aims to quantify the decrease in the pregnancy per artificial insemination (P/AI) in the case of high milk ketones before and after AI. A total of 388,731 test-day from the Brittany Milk Recording Program in France from 226,429 cow-lactations were provided for this trial. For each test-day, information about lactation characteristics, date of AI, date of the following calving, and acetone and β-hydroxybutyrate (BHB) values were included. Ketones were predicted by Fourier transform mid-infrared spectroscopy using MilkoScan Foss analyzers (Foss, Hillerød, Denmark). Many thresholds were evaluated to define cows with hyperketolactia. Hyperketolactia statuses were then categorized into 1 of 4 possible classes according to the milk ketone dynamics for each AI and each threshold of acetone or BHB values (low-low, high-low, low-high, and high-high) within 20 d before and after AI. Similarly, the dynamics of udder health were characterized by changes in somatic cell counts measured at the same test day as ketone bodies. A logistic regression with a Poisson correction was performed to explain the relationship of P/AI with milk ketones and somatic cell count dynamics. Predicted acetone and BHB ranged from -0.51 to 4.92 mM (mean = 0.08 mM, SD = 0.10 mM) and -0.62 to 5.85 mM (mean = 0.07 mM, SD = 0.1 mM), respectively. Hyperketolactia defined by high acetone levels before AI was not associated with decreased P/AI, but high acetone levels after AI were associated with a >10% reduction in P/AI for all thresholds >0.10 mM. Hyperketolactia, defined by high BHB values before, after, or before and after AI, was associated with a 6 to 14% reduction in P/AI compared with cows with low BHB values. These associations are lower than those reported in previous trials in which blood ketones were used. High ketones in advanced lactation are likely to be the result of various primary disorders (secondary ketosis). Because the present work demonstrated that this situation is considered a risk factor for deteriorated reproductive performance, we suggest that high ketones in early and advanced lactation should be of interest to farm advisors.