Skip to main content
NCBI home page
Animal Frontiers: The Review Magazine of Animal Agriculture logoLink to Animal Frontiers: The Review Magazine of Animal Agriculture
. 2026 Feb 15;16(2):86–98. doi: 10.1093/af/vfag007

Present and future opportunities for artificial intelligence applications in dairy cattle reproduction

Julio O Giordano 1,✉, Ana L Laplacette 2
PMCID: PMC13171239  PMID: 42137897

Implications.

  • Artificial intelligence-driven tools could transform reproductive monitoring and management.

  • A full reproductive cycle of cattle requires completion of several biological processes and events such as estrus, ovulation, pregnancy, and parturition that could be monitored and predicted by Artificial intelligence tools. Reproductive monitoring and management tasks and decision-making could also be enhanced or fully executed by Artificial intelligence tools.

  • Artificial intelligence-driven tools for reproductive monitoring and management are emerging, with most at a nascent stage and created based on machine learning algorithms.

  • Development of effective AI-driven tools requires a better understanding of animal biology through data and the consequences of management and environmental factors on data used by Artificial intelligence models.

Introduction

Successful reproductive monitoring and management of cattle is the result of a complex interplay between animal biology and execution of several monitoring and management tasks. The reproductive cycle of female cattle involves physiological and endocrine events enabling conception, gestation, and parturition, after which lactation begins. Parturition, recovery of uterine health, return to normal ovarian cyclicity, expression of estrous behavior, ovulation, and pregnancy are the result of a series cyclical endocrine, behavioral, and physiological patterns. Monitoring and managing female cattle through these complex biological processes requires completion of several tasks, interventions, and decisions usually executed by personnel with technical knowledge and a specific skillset. The diversity, complexity, and temporal dynamics of events and interventions during the reproductive cycle coupled with advances in automation and data-driven technology has primed the field of reproduction for disruptive artificial intelligence (AI) tools. Although most tools for AI-driven decision-making and semi- or full automation of management tasks including detection of estrus, synchronization of ovulation, reproductive program selection, and cow reproductive status monitoring are at a nascent stage or under development, these tools are likely to transform reproduction in dairy farms.

As for most disciplines in the animal sciences, the field of cattle reproduction has traditionally relied on conventional statistical methods. Mathematical models including linear and generalized linear models, mixed effect models, and time-to-event models were used to explore and test relationships between variables of interest and infer or predict outcomes of interest. This approach provides easily interpretable results, but major limitations included the inability to process large and complex datasets with diverse types of data (e.g., time-series and cross-sectional variables, imaging data) or handle nonlinear relationships between variables. These limitations coupled with the expectation of improved performance of AI models and a surge of data-driven technologies for farming have fueled unprecedented growth in the exploration of AI-driven tools for cattle reproduction. Among other reasons, AI is expected to outperform traditional methods because of its flexibility to work with diverse predictors in “big data”, handle complex relationships in the data, and ignore a priori assumptions of relationship between variables (Sarker, 2021; Heil et al., 2023; Ratkovic, 2023).

AI is a field of computer science that enables machines to mimic human intelligence through data-driven methods that use observations as examples to learn how to understand, classify, or make predictions from data. Machine learning is a branch of AI that consists of providing machines with the ability to automatically learn without being explicitly programmed (Samuel, 1959). Although different ways to categorize AI methods exist, machine learning algorithms (MLA) are typically classified as supervised, unsupervised, and reinforcement learning. Supervised learning models use labeled data to learn and tend to be used for classification or regression tasks (Ayodele, 2010). This type of algorithm has been the most explored for application in cattle reproduction. Linear classifiers, logistic regression (LR), support vector machine (SVM), decision trees (DT), random forest (RF), boosting, neural networks (NN), and Bayesian networks (BN) are examples of the most popular supervised MLA used for classification tasks (e.g., estrus vs. no estrus, pregnant vs. nonpregnant) in reproduction. This group of algorithms could also be classified into two main groups. Classic or nondeep MLA rely on human participation to perform feature engineering and to signal differences between variables. On the other hand, deep learning is a subfield of ML based on artificial neural networks. These networks include multiple layers that enable the algorithm to deal with increasingly complex and larger amounts of data such as the time series, high frequency data generated by wearables and nonwearables sensors for cattle (e.g., activity, temperature, rumination, milk yield, body weight sensors). Deep learning models can automatically learn features and interactions between variables even from unstructured data at the expense of requiring large amounts of data to train the networks. More recently, a growing number of AI methods, including convolutional neural networks (CNN) and large language models (LLM) began to be explored for tasks including interpretation of image data such as ultrasonography scans, image-based detection of reproductive events, and dairy records mining (Andrade et al., 2023; Wang et al., 2024; Gontijo et al., 2025). Although these forms of AI might be more powerful, little is known about their value in reproductive monitoring and management because limited research has been conducted with these relatively novel methods.

The goal of this review is to explore some of the key opportunities for application of AI-driven tools in dairy cattle reproductive monitoring and management in commercial dairy farming systems using traditional management practices. The review focuses on the type and characteristics of data and tools that will enable advances in reproduction in dairy systems. Special emphasis is placed on research that has already explored and paved the way for using AI-driven methods to monitor reproductive events and status, and to predict the reproductive potential of cows. Finally, challenges encountered during development of AI tools and their subsequent on-farm implementation are discussed.

Enabling Data and Tools for AI-driven Reproductive Monitoring and Management

The opportunities to develop AI-driven tools for reproduction are growing rapidly as the number, diversity, and complexity of technologies available and data streams generated in dairy farms expand and diversify. Numerous wearable and nonwearable sensors, herd management software tools, genomics, and environmental monitors are or will soon be available for commercial application. Throughout the reproductive cycle, wearable and nonwearable sensors capture a myriad of cow features, behaviors, physiological, and performance parameters valuable for monitoring reproductive events (e.g., calving, estrus) and status (e.g., pregnancy) of cows (Figure 1) or predicting reproductive potential (Figure 2). A deep phenotype emerges from the vast amount of data generated at high frequency and volume from cows, herds, people, and the environment.

Figure 1.

For image description, please refer to the figure legend and surrounding text.

Open in a new tab

Graphical depiction of sensor and non-sensor technologies and data used for development and implementation of AI-driven tools for monitoring reproductive events and status through the reproductive cycle of dairy cows and heifers. Numerous behavioral, physiological, and performance parameters and features can be combined with herd management and performance records and environmental conditions data to generate AI-driven alerts that help personnel or equipment conduct herd monitoring and management tasks including detection of calving, estrus, pregnancy, and health disorder events.

Figure 2.

For image description, please refer to the figure legend and surrounding text.

Open in a new tab

Graphical depiction of sensor and non-sensor technologies and data used for development and implementation of AI-driven tools for predicting the reproductive potential of dairy cows throughout a reproductive cycle. Cow genomic, behavioral, physiological, and performance parameters and features can be combined with herd management and performance records and environmental conditions data to generate AI-driven predictions to inform cow and herd management interventions and decision-making. Predictions of estrus expression and conception risk, overall reproductive potential, the risk of pregnancy loss, and profitability metrics can be estimated and predicted for informing interventions depending on the stage of the reproductive cycle. AIE, artificial insemination at detected estrus; TAI, timed artificial insemination; VWP, voluntary waiting period.

Unlike point-in-time analyses, tests, or records, which only offer a glimpse of physiological conditions and events, fluctuations in patterns of highly dimensional sensor data collected every seconds, days, or weeks more closely reflect the unique behavioral and physiological changes experienced by cows during reproductive events and different stages of the reproductive cycle. Ultimately, the combination of sensor and nonsensor data can generate phenotypes that evolve dynamically over the reproductive cycle as information is learned about new events, and more data is generated at high frequency and volume. For example, the phenotype available for predicting calving only includes cow features, previous reproductive cycle records, and behavioral data for the days and hours immediately before calving. In contrast, the phenotype available for predicting pregnancy to first insemination can combine data collected until calving and the entire voluntary waiting period (VWP). The latter not only captures more biological variability over a longer period but also contains a larger and more diverse group of features with potential predictive value of cow reproductive potential.

Opportunities for AI-Driven Reproductive Monitoring and Management

Every task and decision-making point in the reproductive cycle of cattle presents an opportunity for implementation of AI-driven tools based on the deep phenotype generated from the multiple sources of high dimensional data produced by data-driven tools and monitoring devices. These could contribute to automate, facilitate, and improve routine tasks and decisions between calvings for cows and from birth to first calving in heifers. Augmenting labor or reducing the manual workload for personnel conducting these tasks can streamline management, fill a growing skilled labor gap, and reduce the burden of interventions on cows. Furthermore, predicting reproductive events and outcomes can contribute directly or indirectly to improving herd management practices and performance. Biological and management differences between heifers and cows including physiological responses to lactation, the availability of milk data for cows and not for heifers, and different types of facilities and level of investment on data-driven technologies for heifers versus cows present both challenges and opportunities for development and implementation of AI tools specifically for heifers and cows.

Starting at the peri-calving period, AI models based on behavioral, physiological, and computer vision (CV) data could be used to predict the time of calving within a timeframe that enables moving cows to calving pens at an appropriate time. This could eliminate frequent visual observation of impeding signs of calving by personnel. Models using the same or additional data could also predict dystocia events and monitor calving progression, facilitating interventions to improve calving outcomes.

Poor health in early lactation is strongly associated with reproductive performance (Bruinjé et al., 2023). Therefore, predicting and monitoring health events after calving could help identify cows at greater risk of reproductive failure and enable implementation of strategies to mitigate the effects of poor health on reproductive outcomes.

The VWP, which encompasses the time from calving until cows become eligible for pregnancy, provides ample opportunities for implementing AI tools. Predictive models based on data collected during this period aid in decision-making. For example, automating monitoring of reproductive physiological status could enable prescriptive models that assign cows to targeted reproductive management (TRM) strategies (Giordano et al., 2022). Deep phenotyping of recovery of uterine health, return to ovarian cyclicity and estrus, and metabolic adaptations to lactation could be used individually or combined with herd management features and environmental conditions data to develop prescriptive models that select effective management strategies for a herd. Defining an ideal duration of the VWP and timing of pregnancy for all or subgroups of cows are difficult decisions with important consequences to herd performance, and profitability. Models integrating data collected through the VWP could be used to predict an optimal VWP duration and timing of pregnancy that optimizes reproductive performance and profitability. For example, an AI tool could combine data for return to estrus after calving (i.e., timing of first estrus and number of estrus events) to predict the likelihood of pregnancy if inseminated with projections of profitability based on milk yield and lactation length to decide if insemination upon detection of estrus is convenient or should be delayed.

After the end of the VWP, AI tools could aid monitoring and management tasks such as detection of estrus and selection of synchronization of ovulation protocols to optimize and simplify submission of cows to insemination or embryo transfer. Although there is ample evidence of the accuracy of non-AI models to identify cows in estrus based on behavioral parameters like physical activity (Mayo et al., 2019; Schilkowsky et al., 2021), incorporating other parameters into AI models could help improve monitoring. Models that predict the quality of estrus based on behavioral features (e.g., intensity, duration, rate of activity increase) would be valuable for decisions at the time of insemination or thereafter. Automated endocrine and behavioral profiling (i.e., serial determination of hormone levels and behavioral parameters like activity, resting, rumination, and eating behavior) could be used to estimate the expected fertility of a cow detected in estrus by an automated estrus detection system.

As the options for generating offspring from dairy cattle expand, opportunities to use AI driven tools at or around the time of offspring generation also increases. AI-driven decision-support systems could be developed to choose from the variety of methods to generate a pregnancy. Natural services, artificial insemination with conventional or sexed semen, using frozen or fresh semen, or embryo transfer are available options for dairy farms. The type of offspring could also be chosen by AI tools that optimize genetic progress and the ratio of females born for replacement versus value added cross-bred or beef-breed calves. For example, a model could determine the best opportunity to generate a female dairy replacement using sexed frozen semen based on the genetic makeup of the dam and available sires accounting for the cow predicted probability of pregnancy in combination with data for herd structure, future replacements, and offspring value.

Collecting data until the time of and immediately after pregnancy generation through artificial insemination or embryo transfer can also provide opportunities for implementation of AI-driven interventions. Automated endocrine profiling could support implementation of hormonal and nonhormonal therapy before and after pregnancy generation. Targeted nutritional strategies that could enhance or support embryo development could be implemented through individual cow or group feeding. Hormonal supplementation (e.g., progesterone) could be provided to subgroups of cows through exogenous hormone administration (e.g., injections or intravaginal delivery) or endogenous production through manipulation of ovarian function (i.e., accessory corpus luteum formation).

Automated or semi-automated pregnancy status monitoring through in-line milk hormone profiling systems, cow-side hormone tests, behavioral and physiological sensing, and ultrasonography is another major area of opportunity. AI-driven tools could detect and longitudinally monitor pregnancy establishment and loss or determine pregnancy status at discrete time points while minimizing or eliminating challenges of traditional pregnancy testing methods. At nonpregnancy diagnosis, predictive models could identify optimal re-insemination strategies.

Monitoring Reproductive Events and Status

A rich area of application of AI-enabled tools is the detection of events that trigger reproductive management interventions (Figure 1). Calving and estrus events lead to immediate actions whereas confirmation of pregnancy status triggers and informs interventions in the short and long-term. Calving triggers a move to a maternity pen and estrus triggers immediate insemination. Pregnancy confirmation triggers immediate actions such as initiation or completion of a resynchronization of ovulation protocol or longer-term actions like pen moves, vaccinations, or wearable sensor removal.

Calving prediction tools can help farm personnel move cows to calving pens whereas calving monitoring tools could help monitor calving progression. Borchers et al. (2017) trained three MLA (RF, Linear Discriminant Analysis, and NN) to predict calving time using activity, lying, and rumination behaviors collected by neck- and leg-attached sensors. The ability to predict an 8-h period before calving on the day of calving with algorithms that combined all behaviors was reasonable (Sensitivity [Se]: 72% to 83%, Specificity [Sp]: 75% to 80%; Positive Predictive Value [PPV]: 61% to 69%; Negative Predictive Value [NPV]: 85% to 90%). Higaki et al. (2020) also reported that a SVM model trained on 15 features created from continuous measurements of ventral tail base skin temperature and the expected calving date, had Se of 82% to 85% and PPV between 68% and 72%. Although in both studies results seemed favorable, achieving the highest Se possible is critical for calving prediction to avoid cows calving in the wrong place if not moved to calving pens. Likewise, specificity is of outmost importance because false positive alerts could result in moving cows to calving pens unnecessarily or too early. Thus, despite the apparent high performance of some of the algorithms developed based on behavior or body temperature, additional research is necessary to develop AI-driven tools that approach 100% Se and Sp for predicting calving, if these are to be widely used in dairy farms. Computer vision is a promising technology for monitoring calving but has not yet been extensively developed. The obvious changes in behavior and physical appearance of cows immediately before and during calving are detectable through video monitoring. McDonagh et al. (2021) reported that a neural network model trained on images of multiple behaviors including standing, lying, walking, shuffling, eating, drinking, and abdominal contractions correctly identified these outcomes ∼80% of the time. However, the algorithm was not tested for predicting and alerting calving. Developing CV systems for calving prediction for commercial application poses significant challenges including occlusion, poor lighting, individualizing cows, and several others.

Detection of estrous behavior has been successfully automated for more than 30 years. Today, a myriad of automated estrus detection systems based on wearable sensors that monitor behavioral and physiological parameters are available for commercial farms (Mayo et al., 2019; Schilkowsky et al., 2021). The dramatic change in physical activity associated with estrus is so well defined and of such magnitude in most cows that acceptable estrus detection accuracy has been achieved with estrus alerts created without AI. Although this might explain less interest in developing AI-driven algorithms, there are still opportunities to improve estrus detection efficiency and accuracy. Combining multiple behaviors and physiological parameters with cow, herd management, and environmental data could help minimize or eliminate current limitations. For example, insufficient increases in activity in some cows result in false negative outcomes, whereas management interventions such as pen moves or other actions that cause a rise in activity increase the rate of false positive alerts. Parameters that change during estrus like rumination and eating behavior, body temperature, and milk yield are candidates for inclusion into estrus detection algorithms (Dolecheck et al., 2015). Nonsensor data such as pen moves, vaccinations, hoof trimming or any other event that could alter activity patterns might also be of value. Recently, AI-driven estrus detection algorithms including cow location (Wang et al., 2020), multiple engineered acceleration features (Shahriar et al., 2016; Wang et al., 2020), and vaginal temperature and conductivity (Higaki et al., 2019) have been reported. A plethora of MLA including supervised and nonsupervised classifiers and deep learning algorithms were trained and tested (Shahriar et al., 2016; Higaki et al., 2019; Wang et al., 2020). Although this new algorithms were effective (Shahriar et al., 2016; Higaki et al., 2019; Wang et al., 2020), the range of performance values was within that observed for many of the traditional non-AI driven algorithms (Mayo et al., 2019). Most studies testing novel AI algorithms had small sample sizes and were conducted in controlled research settings. Thus, future research should focus on demonstrating their value with large sample sizes and under commercial farm conditions.

CV systems could represent a major paradigm shift in detection of estrus as this technology could either replace or complement wearable sensors. Wang et al. (2024) recently demonstrated feasibility for detecting cows in estrus based on a CV system that used the You Only Look Once V8 Nano (Yolo-v8n) model. The improved model called Estrus-YOLO was trained to detect cow mounting and standing behavior and label individual cows housed in a free-stall barn with access to exercise lots with dirt flooring. The reported ability to detect cows in estrus based on standing while mounted was above 90% which suggested potential for application. This and most other studies that explored CV for detection of estrus (Tsai and Huang, 2014; Guo et al., 2019) were conducted in research settings, under controlled conditions, with limited number of animals, and for relatively short periods of time. Thus, it remains to be determined whether advances in CV will overcome challenges in commercial farms including poor lighting, similarities among cows, high stoking density in confined facilities, and regular regrouping of cows. Technical issues including inadequate internet connectivity, faulty data transmission, and frequent power loss must also be resolved for effective implementation in commercial farms.

Pregnancy status determination and detection of pregnancy loss are key management tasks. Thus, developing AI-tools to diagnose pregnancy using noninvasive data-driven technologies is of considerable value. To this end, the milk infrared (MIR) spectra have been used to train MLA to predict pregnancy status based on changes in spectral features associated with pregnancy. Delhez et al. (2020) trained PLS-DA models using milk spectral differences from monthly milk tests. Models trained on data recorded up to 150 d after insemination had the highest performance (AUC = 0.76), although lower than needed for on-farm implementation. Similarly, Brand et al. (2021) used deep learning to predict pregnancy status from milk MIR spectral data. The best-performing model had an accuracy of 88%, sensitivity of 89% and specificity of 86% suggesting that deep-learning models trained on observed milk MIR spectra could be a useful tool to predict pregnancy status. Although this approach to pregnancy prediction might not replace traditional methods as acceptable performance at early stages of pregnancies has not been achieved, milk spectra-based testing has been proposed for supporting other routine practices such as identifying pregnancy losses or verifying conception dates (Brand et al., 2021). Although some of the pregnancy status and loss models trained on milk spectral data generated reasonable results, results were suboptimal early in gestation to support aggressive reproductive programs. Seeking novel methods and other behavioral and physiological data affected by biological pregnancy signals to feed into MLA might lead to the development of more effective tools.

Predicting Reproductive Potential

Identifying cows with different reproductive potential in advance of key steps of the reproductive cycle or execution of management tasks is valuable for reproductive management (Figure 2). For example, predicting the likelihood of expressing estrus after the end of the VWP, the probability of pregnancy to an insemination or embryo transfer, and the effects of timing of pregnancy on profitability can enable implementation of TRM for individual or groups of cows with different reproductive potential (Giordano et al., 2022). Through novel on-farm technologies and developments in the field of AI, the type, variety, and amount of data and methods to predict relevant reproductive outcomes have evolved rapidly (Figure 2). Dairy farms have access to genomically enhanced predictions of fertility and detailed high frequency data from sensors that capture complex associations among genetics, reproductive behaviors, and physiological outcomes. Also common in dairy operations are herd management and performance records software and environmental conditions monitoring tools. Collectively, access to novel data streams and AI-methods is paving the way for major leaps in development of predictive and prescriptive AI-driven tools for interventions before and after insemination, interventions after pregnancy confirmation, and through the entire lactation (Figure 2). Developing AI-driven models to identify individuals or groups of cows with different reproductive potential is a logical goal given the wealth of evidence of associations between genetics, behavioral, physiological, and performance parameters with cow reproductive potential (Giordano et al., 2022; Rial et al., 2022; Bruinjé et al., 2023; Mendonça et al., 2025). The expectation is that AI models can be more effective than traditional modeling at extracting information and classifying cows based on multiple features due to multiple reasons. These include complex associations between drivers of biological processes and reproductive outcomes, significant biological overlap among drivers of fertility, the temporal dynamics presented by some predictors, and the insufficient predictive value of certain associations when used individually.

Earlier research efforts focused primarily on training MLA to predict the outcome of inseminations using herd- and cow-level data only because data from wearable and nonwearable sensor systems or genomics were not readily available. More recently, a growing number and variety of cow level genotypic and phenotypic data is being used to train MLA. In an early attempt to use MLA to predict the outcome of inseminations, Shahinfar et al. (2014) trained five classifier MLA (naive Bayes, BN, DT, bagging, and RF) with up to 25 production, reproduction, health, and genetic herd- and cow features from 26 farms that milked Holstein cows in a variety of housing systems. The best predictions were for models based on an RF algorithm (AUC of 0.76 and 0.74 for primiparous and multiparous cows, respectively). Herd-year-month of insemination and herd-level pregnancies per AI, and individual cows days in milk and number of inseminations, and health events data were the most informative. Although relatively superior to most models subsequently reported in the literature, the performance metrics reported by Shahinfar et al. (2014) were for models evaluated only through cross-validation rather than using hold-out datasets and did not follow best ML evaluation practices including train-test split methods that avoid overfitting, temporal validation, and cross-farm generalization.

Subsequently, attempts to incorporate other information from individual cows to enhance predictions were made. In this regard, MIR spectra is one of the most thoroughly evaluated predictors of reproductive potential because of the widespread availability of spectral data generated through milk testing programs. Hempstalk et al. (2015) predicted the likelihood of conception to an insemination with eight MLA (C4.5 DT, naive Bayes, BN, LR, SVM, PLS regression, RF, and rotation forest) trained on data for multiple herd- and cow-level factors and MIR spectra. Although not consistent across all scenarios, LR models had best-performance with AUC-ROC from 0.50 to 0.62. Contrary to expectations, inclusion of milk MIR spectral data did not contribute substantially to the predictive value of the models. More recently, Lou et al. (2024) trained multiple MLA models to identify groups of cows with high versus low fertility using a combination of cow-level information, milk yield and component, and other milk features extracted from MIR spectra. The AUC-ROC in herd-by-herd validation of LR, PLS-DA, and RF models was as low as 0.57 for RF and PLS-DA models and as high as 0.77 for the best performing PLS-DA model. These results were a substantial improvement compared with Hempstalk et al. (2015) and suggested there is value of MIR data for prediction of pregnancy potential. As more on-farm milk analyzer technologies become available there will be more opportunities to explore the use of milk spectra and milk features to predict pregnancy.

As behavioral, physiological, and performance data from automated livestock monitoring systems (ALMS) using wearable and nonwearable sensors became more readily available, the inclusion of these data into predictive models based on MLA has been recently explored. Information extracted from these systems has been extensively associated with the likelihood of reproductive success (Giordano et al., 2022; Rial et al., 2022; Bruinjé et al., 2023; Mendonça et al., 2025). Mounting evidence has demonstrated that records of estrus and health disorder alerts as well as features from the pattern of several individual parameters (i.e., physical activity, rumination, eating behavior) recorded by ALMS during the VWP can help identify cows more likely to express estrus and become pregnant.

To explore the value of high frequency and multidimensional ALMS data for predicting the outcome of the first insemination postpartum in lactating dairy cows, we used data from ALMS from a single commercial herd in the United States to train several MLA models (Granados et al., 2022). In total, data from 3,002 Holstein cow lactations with 139 features from several behaviors (activity, resting, eating, and rumination), intraruminal temperature, body weight, and milk features (milk yield, fat, protein, and lactose in milk, and conductivity) collected daily up to 1 day before insemination and summarized in multiple or single periods were used to train several classifier MLA. The best models had accuracy in the range of 54% to 62%, which was relatively low. These results suggested that including many features from ALMS into MLA models may not dramatically improve the ability to predict pregnancy compared with other types of data. However, these results should be interpreted with caution because of a limited number of observations available in the dataset, potential masking of reproductive potential differences among cows due to the use of timed insemination with a fertility program, and no attempts to use more complex methods to improve the performance of models. Notwithstanding, results from Granados et al. (2022) agree with more recent results from Hannon et al. (2025), who used sensor data from ALMS and nonsensor data from 1,130 cow lactations from 13 herds with automated milking in the United Kingdom and Ireland to predict the likelihood of expression of estrus after the end of the VWP and pregnancy to first service. Several raw or engineered features from daily data for milk parameters including yield, components, and conductivity, robot visits, feed intake in the robot, and rumination and activity combined with other cow features and performance data were used to train RF models. The AUC for the ROC for the best models including only robotic milking system data or the same data combined with wearable sensors and cow features were 0.6 to 0.65, respectively for predicting expression of estrus, and 0.56 and 0.62 for predicting conception success when evaluated in the test dataset. Although not directly comparable to results from other MLA models using more herd-level and MIR data, models incorporating high-frequency ALMS data have shown comparatively lower predictive performance. Only Marques et al. (2024) reported substantially better results for models that predicted the likelihood of pregnancy using data from ALMS. The best-performing LR, linear discriminant, and SVM algorithms had AUC-ROC in the range of 0.86 to 0.87. However, it is hard to determine if the high predictive ability of these MLA was due to the ability of the algorithms to extract valuable information from the data or the result of including information available close to insemination. Specifically, the algorithms tested included rumination time, automated estrus alerts, and estrus intensity, all of which are strongly associated with the success of TAI services (Laplacette and Giordano, 2025) but become available only within hours before insemination. Unlike the earlier decisions potentially enabled by AI tools developed with data from weeks or several days before insemination, algorithms including data available on the day of or a few hours before AI are only useful for decision-making at the time of or after insemination. For example, such tools could be used to select the type of semen used for insemination or implement a post-insemination hormonal therapy but would not enable selecting the protocol for synchronization of estrus or ovulation.

To the best of the authors knowledge, most predictive models developed for identifying cows of different reproductive potential reported in the literature have not yet been deployed to farms or their performance has not been evaluated prospectively in commercial farms. There are two examples of fertility prediction models developed using MLA that were tested prospectively in commercial farms. However, these algorithms are proprietary, and the methods used to create the models have not been fully disclosed. Sitko et al. (2023) reported that primiparous Holstein cows from 6 commercial farms grouped in high-, medium-, and low-fertility groups based on a “reproduction index” that predicted days to pregnancy based on MLA presented meaningful statistically significant differences for several reproductive outcomes. Moreover, cows in different fertility groups presented a different response to programs that prioritized insemination at detected estrus or timed insemination. The algorithm used for grouping cows included gradient boosting machine, RF, and SVM, and was trained with data from 45 gPTA (genomic predicted transmitting ability) traits and haplotypes from the CDCB (Council for Dairy Cattle Breeding) and wellness traits from a company product (Clarifide, Zoetis). Mendonça et al. (2025) also demonstrated that a “fertility index” created based on an algorithm trained on behaviors such as activity, rumination, eating, and other lifecycle production features successfully separated cows with different reproductive potential. No information was provided on the methods used to create this proprietary algorithm (Merck Animal Health). These two examples demonstrated the feasibility of implementing AI tools for separating cows with different future reproductive potential for used in TRM. A caveat of these two algorithms is that the potential predictive power may not have been maximized as genomic and ALMS data were not combined. Combining these type of data into MLA models to predict reproductive potential is an active area of research. For example, we recently reported the results of models to predict the outcome of first service using a combination of behavioral and performance parameters from ALMS, genomic predictions for fertility, and several other cow features (Laplacette and Giordano, 2025). Although predictions for individual cows were not as high as expected (i.e., all AUC-ROC <65%), grouping cows in tertiles of predicted probability of pregnancy from the best MLA models resulted in meaningful differences in first service P/AI (e.g., 24% to 26% points between the low- and high-fertility group for an SVM model). Developing, testing, and deploying AI-driven tools based on MLA trained on diverse datasets is an active and rapidly evolving area of research in dairy cattle reproduction.

Several observations emerge from the available literature. Notably, there are no well-established and accepted methods to select MLA for development of predictive models. Except for LR, which has been used in most studies, and the use of PLS when using MIR data, there is no consistent use of MLA in research studies. This lack of consistency not only hampers comparisons across studies and scenarios of MLA use but also complicates efforts to identify most appropriate or effective MLA, model training and validation methods, and the type of data needed to successfully predict reproductive potential. This issue is becoming increasingly complex as the diversity and number of MLA, and the type and amount of data available to researchers expands. A common theme in research studies has been to test several MLA concomitantly to identify the most predictive type of algorithm. Tools designed to simplify evaluation of many MLA concomitantly such as auto ML software packages were explored for other applications in dairy production (Perez et al., 2025). However, software package limitations and the limited control that researchers have over certain functions of these tools can limit the value of this approach. Another observation from the literature is that most of the existing research has been confined to nondeep learning classifier MLA. Research in dairy cattle reproduction with other novel and powerful AI methods including deep learning, LLM, and other new forms of AI is either nonexistent or at its nascent stages. This is likely due to a combination of factors. Primarily the fact that many of these AI methods are still relatively new, whereas for others lack sufficient data prevented their widespread exploration.

Past research has also shown that despite having multiple sources of detailed high frequency and single point data for the myriad of biological, management, and environmental factors that drive or are associated with reproductive potential, it is hard to predict outcomes of biological events and management interventions for individual cows (e.g., predict if a cow will express estrus, become pregnant if inseminated). Although breakthroughs in AI methods and new sources of data could soon resolve this problem, grouping cows based on thresholds of predicted probabilities scores from MLA models is a compromise with potential value for practical application of AI-driven tools (Figure 3). Thresholding consists of grouping cows based on selected cut points of individual cow predicted probabilities from models that provide probability scores. These cut points can be selected based on needs and opportunities to improve herd performance and management practices. As many groups of cows as desired or necessary can be created, provided that expected differences between groups are meaningful to justify implementation of TRM strategies. For example, cows within a herd or cohort could be split into thirds of predicted probability of pregnancy to generate groups of high, medium, and low-fertility potential. If the original model has predictive power, meaningful differences in outcomes such as pregnancies per AI should be observed.

Figure 3.

For image description, please refer to the figure legend and surrounding text.

Open in a new tab

Conceptual framework for development and implementation of AI tools to predict the reproductive potential of lactating dairy cows and create groups for targeted reproductive management. Major steps in the process include obtaining and processing data for cow and herd features, behavioral, physiological and performance parameters features and events, and environmental conditions collected from different sources including genomics and wearable and non-wearable sensors. Machine learning models are trained to predict the probability of an event or outcome of interest (e.g., estrus, pregnancy, pregnancy loss) for individual cows. Cutoff values for predicted probability scores from MLA are used to create groups of cows with meaningful differences in reproductive potential for implementation of targeted reproductive management strategies. For example, cows can be separated into groups with expected high, medium, and low pregnancies per AI based on the probability score generated by a machine learning algorithm model. MLA, machine learning algorithm; P/AI, pregnancies per AI.

Use of Artificial Intelligence in Other Areas of Reproduction

This review did not cover several other possible areas of reproductive monitoring and management primmed for application of AI tools. Using MLA for different aspects of assisted reproductive technologies (ART) for cattle such as oocyte (Raes et al., 2025), semen (Hürland et al., 2023), embryo (Angel-Velez et al., 2023), and donor animal selection (Marín-Urías et al., 2023) has also been explored. AI-enabled systems are already available for ART in humans (Abou Ghayda et al., 2024). AI-tools could be implemented and have been explored for semen sorting and analysis (Urli et al., 2025). Algorithms to detect pregnancy and identify reproductive tract and ovarian structures based on transrectal ultrasonography images have been developed and tested (Gonçalves et al., 2025).

Overcoming Challenges to Unleash the Power of AI in Cattle Reproduction

There are several general hurdles to overcome before AI-driven tools can truly revolutionize cattle reproduction.

Large herd-to-herd variability in reproductive management generates numerous challenges. AI tools are more difficult to develop, AI tool performance can be reduced, and tools cannot be easily generalized because of substantial variation across herds and their programs to submit cows to insemination or embryo transfer, pregnancy generation methods (i.e., insemination, embryo transfer, natural service), VWP length, pregnancy testing methods, and the strong influence of decision-makers. Novel methods are needed to adapt AI tools to the vast array of farm management practices and conditions. AI-driven models that can be generalized but also calibrated to accommodate specific farm conditions can contribute to overcoming these issues.

Technical challenges including missing data, inaccurate measurements, variability in reporting across different sources of data, countless AI methods, and lack of fully automated real time data integration tools continue to limit progress and affect AI-tool performance. Several AI and non-AI enabled solutions will be required to address these challenges including novel strategies for data imputation and synthetic data creation, standardization of input data across farms, comprehensive methods to select optimal MLA, and implementation of data integration tools (You et al., 2023; Cabrera et al., 2025; Perez et al., 2025). The latter is critically needed as AI tools are likely to use highly dimensional data from a variety of continuous and point-in-time data streams from cows, facilities, equipment, and the environment. To this end, open communication and data transfer between dairy herd management software or standalone data integration tools with other data generation systems commercialized by multiple companies will be required to enable AI tools. This can be accomplished through collaborations among providers of data generation systems, commercialization of data between parties involved in data generation and use, a more diverse portfolio of data generation systems from individual companies, or combinations thereof.

Lack or limited access to data for scientists at academic institutions or research centers stifles innovation and progress. Because AI tool development requires large and diverse datasets that in many cases are only accessible to companies, many of the methods and models developed are proprietary to companies that commercialize data-driven technologies. Academic researchers are unable to fully contribute to improving, validating, and implementing most of these AI tools. The scientific community and industry must engage in more productive collaborations to hasten the development of effective and safe AI tools.

Lastly, our understanding of how data reflects cow reproductive biology and the influences of management and the environment on cows are still limited. It is imperative to improve our understanding of cow biology through data used or generated by AI-driven tools.

Overcoming all or many of these hurdles is likely to unleash the true power of AI methods for reproduction.

Summary and Conclusions

The opportunities to change reproductive monitoring and management through AI-driven tools are vast and diverse. Cows undergo multiple physiological events and receive interventions to complete a successful reproductive cycle. AI-driven tools can contribute to monitoring and managing cows through these events, automate or semi-automate interventions, and drive or enhance decision-making. Among the vast and growing array of AI methods and tools available, classifier MLA have been used and explored most extensively. AI-driven algorithms have been developed for detection of calving events, estrus, and pregnancy status determination. Likewise, MLA models have been trained to predict cow reproductive potential with varying levels of success. The AI tools developed so far had a wide range of performance and probably low levels of adoption due to limited deployment to farms. A significant amount of work remains to be done to truly revolutionize reproductive monitoring and management with AI-driven tools. Fueled by an ongoing surge in the number and types of data-driven technologies, and rapidly evolving AI methods, the field of cattle reproduction is primed to be disrupted by AI.

Acknowledgments

This manuscript was invited for submission by the American Society of Animal Science. The views expressed in this publication are those of the author(s) and do not necessarily reflect the views or policies of the American Society of Animal Science, the journal, or the publisher.

Conflict of interest statement. The authors declare no real or perceived conflicts of interest.

Author Biographies

graphic file with name vfag007ilf1.jpg

Dr. Julio O. Giordano, DVM, MS, PhD, is a Professor in the Department of Animal Science at Cornell University and an internationally ­recognized leader in dairy cattle biology and precision management systems. He serves as Director of the Dairy Cattle Biology and Management Laboratory, Co-Director of the Cornell Institute for Digital Agriculture (CIDA), and Director of the Cornell Agricultural Systems Testbed (CAST) and Demonstration Site for the Farm of the Future. His research integrates animal biology, farm management, engineering, and data analytics to enhance reproductive performance, health, and profitability of dairy cattle through technology-driven solutions. Dr. Giordano’s research focuses extensively on reproductive biology and management of dairy cows, including hormonal manipulation of the estrous cycle to improve fertility and development of strategies to conduct and improve automated detection of estrus. He has pioneered the development and evaluation of automated monitoring technologies, precision management strategies, and data-driven tools that assess economic impacts of reproductive and health management programs. His innovative approaches combine sensor-based health monitoring, reproductive biotechnologies, genomics, machine learning, and IoT-based systems for transforming dairy production systems and enabling producers to make informed, real-time decisions that enhance efficiency and sustainability. Dr. Giordano earned his Doctor of Veterinary Medicine from the Catholic University of Córdoba in Argentina, Master of Science in Animal Science from the University of Tennessee, and Ph.D. in Dairy Science from the University of Wisconsin–Madison. He has published more than 75 peer-reviewed papers and book chapters, delivered over 270 international and national presentations, workshops, and webinars, and secured more than $20 million in research funding. Beyond research, Dr. Giordano is deeply committed to teaching and mentorship while leading courses in dairy herd management, reproductive physiology, and digital agriculture. Through extensive outreach efforts, he collaborates closely with industry partners to translate scientific findings into practical solutions that advance technology-enabled practices for modern agriculture.

graphic file with name vfag007ilf2.jpg

Dr. Ana L. Laplacette is a PhD student in the Department of Animal Science at Cornell University and a member of the Dairy Cattle Biology and Management Laboratory under the direction of Dr. Julio Giordano. She earned her Doctor of Veterinary Medicine from the University of Buenos Aires, Argentina, and completed her Master of Science in Animal Science at Cornell University. Her research focuses on developing data-driven, targeted management programs to enhance reproductive management practices and performance in dairy herds through advanced analytical approaches. Dr. Laplacette’s work emphasizes the development of machine learning and traditional analytical tools to identify groups of cows with varying reproductive potential, enabling precision management strategies tailored to individual animal needs. Her current research objectives include designing and evaluating innovative programs that optimize first service ­management and reproductive performance by integrating automated estrous detection alerts with strategic synchronization of ovulation protocols. This interdisciplinary approach combines reproductive physiology, data analytics, and precision agriculture technologies to create practical solutions for modern dairy operations. She has published more than 30 peer-reviewed papers and abstracts and delivered more than 20 presentations at scientific conferences. Through her research, Dr. Laplacette aims to bridge the gap between advanced computational methods and on-farm implementation, developing tools that enable dairy producers to make informed decisions about reproductive management.

Contributor Information

Julio O Giordano, Department of Animal Science, Cornell University, Ithaca, NY 14850.

Ana L Laplacette, Department of Animal Science, Cornell University, Ithaca, NY 14850.

Author Contributions

Julio Giordano (Conceptualization, Investigation, Writing original draft, Writing—review & editing) and Ana Laplacette (Conceptualization, Investigation, Writing original draft, Writing—review & editing)

Funding

The authors and research from the authors presented in this manuscript are supported by the USDA National Institute of Food and Agriculture (NIFA) Farm of the Future project no. 2023-77038-38865 and by the National Science Foundation Directorate for Computer and Information Science and Engineering Project IIS-2211941. Work was also supported by the USDA National Institute of Food and Agriculture (Washington, DC), Animal Health Program Project no. 2017-67015-26772, Animal Reproduction program Project no. 2020-08769, Hatch project NYC-2020-21-255, and Multistate project 1021189. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the National Institute of Food and Agriculture (NIFA) or the United States Department of Agriculture (USDA). Research presented in this review was also partially supported by Zoetis Inc, Merck Animal Health, and SCR Engineers.

Literature Cited

  1. Andrade A.C., Silva L.O. d, Souza V.F., Rufino L.M. d A., Silva T.E. d, dos Santos A.D.F., Gomes D. d A., Rodrigues J.P.P.  2023. Identifying pregnancy in cows using ovarian ultrasound images and convolutional neural networks—a proof-of-concept study. Comput. Electron. Agric. 206:107674. doi:10.1016/j.compag.2023.107674 [Google Scholar]
  2. Angel-Velez D., De Coster T., Azari-Dolatabad N., Fernández-Montoro A., Benedetti C., Pavani K., Van Soom A., Bogado Pascottini O., Smits K.  2023. Embryo morphokinetics derived from fresh and vitrified bovine oocytes predict blastocyst development and nuclear abnormalities. Sci. Rep. 13(1):4765. doi:10.1038/s41598-023-31268-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Ayodele T.O.  2010. Types of machine learning algorithms. In Zhang Y., editor. New advances in machine learning.  Intechopen, London, United Kingdom; p. 19–48 [Google Scholar]
  4. Borchers M.R., Chang Y.M., Proudfoot K.L., Wadsworth B.A., Stone A.E., Bewley J.M.  2017. Machine-learning-based calving prediction from activity, lying, and ruminating behaviors in dairy cattle. J. Dairy Sci. 100(7):5664–5674. doi:10.3168/jds.2016-11526. [DOI] [PubMed] [Google Scholar]
  5. Brand W., Wells A.T., Smith S.L., Denholm S.J., Wall E., Coffey M.P.  2021. Predicting pregnancy status from mid-infrared spectroscopy in dairy cow milk using deep learning. J. Dairy Sci. 104(4):4980–4990. doi:10.3168/jds.2020-18367 [DOI] [PubMed] [Google Scholar]
  6. Bruinjé T.C., Morrison E.I., Ribeiro E.S., Renaud D.L., Couto Serrenho R., LeBlanc S.J.  2023. Postpartum health is associated with detection of estrus by activity monitors and reproductive performance in dairy cows. J. Dairy Sci. 106(12):9451–9473. doi:10.3168/jds.2023-23268 [DOI] [PubMed] [Google Scholar]
  7. Cabrera V.E., Bewley J., Breunig M., Breunig T., Cooley W., De Vries A., Fourdraine R., Giordano J.O., Gong Y., Greenfield R.  et al.  2025. Data integration and analytics in the dairy industry: challenges and pathways forward. Animals. 15(3):329. 10.3390/ani15030329 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Delhez P., Ho P.N., Gengler N., Soyeurt H., Pryce J.E.  2020. Diagnosing the pregnancy status of dairy cows: How useful is milk mid-infrared spectroscopy?  J. Dairy Sci. 103(4):3264–3274. doi.org/10.3168/jds.2019-17473. [DOI] [PubMed] [Google Scholar]
  9. Dolecheck K.A., Silvia W.J., Heersche G., Chang Y.M., Ray D.L., Stone A.E., Wadsworth B.A., Bewley J.M.  2015. Behavioral and physiological changes around estrus events identified using multiple automated monitoring technologies. J. Dairy Sci. 98(12):8723–8731. doi:10.3168/jds.2015-9645 [DOI] [PubMed] [Google Scholar]
  10. Giordano J.O., Sitko E.M., Rial C., Pérez M.M., Granados G.E.  2022. Symposium review: Use of multiple biological, management, and performance data for the design of targeted reproductive management strategies for dairy cows. J. Dairy Sci. 105(5):4669–4678. doi: 10.3168/jds.2021-21476 [DOI] [PubMed] [Google Scholar]
  11. Ghayda R.A., Cannarella R., Calogero A.E., Shah R., Rambhatla A., Zohdy W., Kavoussi P., Avidor-Reiss T., Boitrelle F., Mostafa T.  ,  et al. ; Global Andrology Forum. 2024. Artificial intelligence in andrology: from semen analysis to image diagnostics. World J. Mens. Health. 42(1):39–61. doi:10.5534/wjmh.230050 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Gonçalves L.M., Fontes P.L.P., Alves A.A.C.  2025. Computer vision ­analysis of luteal color Doppler ultrasonography for early and automated pregnancy diagnosis in Bos taurus beef cows. J. Anim. Sci. 103:skaf166. doi.org/10.1093/jas/skaf166 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Gontijo D., Santana D.R., Costa G.d.A., Cabrera V.E., Freitas E.N.d.A.  2025. Dairy GPT: empowering dairy farmers to interact with numerical databases through natural language conversations. Smart Agric. Technol. 12:101097. doi.org/10.1016/j.atech.2025.101097. [Google Scholar]
  14. Granados G.E., Perez M.M., Giordano J.O.  2022. Prediction of pregnancy in lactating dairy cows with machine learning algorithms using behavioral, physiological, and performance sensor data and other cow, herd, and environmental data. J. Dairy Sci. 105 (E-Suppl. 1) [Google Scholar]
  15. Guo Y., Zhang Z., He D., Niu J., Tan Y.  2019. Detection of cow mounting behavior using region geometry and optical flow characteristics. Comput. Electron. Agric. 163:104828. doi:10.1016/j.compag.2019.05.037 [Google Scholar]
  16. Hannon F.P., Green M.J., O’Grady L., Hudson C., Gouw A., Randall L.V.  2025. Predictive models for the implementation of targeted reproductive management in multiparous cows on automatic milking systems. J. Dairy Sci. 108(2):1634–1643. doi:10.3168/jds.2024-24920 [DOI] [PubMed] [Google Scholar]
  17. Heil B.J., Crawford J., Greene C.S.  2023. The effect of non-linear signal in classification problems using gene expression. PLoS Comput. Biol. 19(3):e1010984. doi:10.1371/journal.pcbi.1010984 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hempstalk K., McParland S., Berry D.P.  2015. Machine learning algorithms for the prediction of conception success to a given insemination in lactating dairy cows. J. Dairy Sci. 98(8):5262–5273. doi:10.3168/jds.2014-8984 [DOI] [PubMed] [Google Scholar]
  19. Higaki S., Koyama K., Sasaki Y., Abe K., Honkawa K., Horii Y., Minamino T., Mikurino Y., Okada H., Miwakeichi F.  et al.  2020. Technical note: Calving prediction in dairy cattle based on continuous measurements of ventral tail base skin temperature using supervised machine learning. J. Dairy Sci. 103(9):8535–8540. doi:10.3168/jds.2019-17689 [DOI] [PubMed] [Google Scholar]
  20. Higaki S., Miura R., Suda T., Andersson L.M., Okada H., Zhang Y., Itoh T., Miwakeichi F., Yoshioka K.  2019. Estrous detection by continuous measurements of vaginal temperature and conductivity with supervised machine learning in cattle. Theriogenology. 123:90–99. 10.1016/j.theriogenology.2018.09.038 [DOI] [PubMed] [Google Scholar]
  21. Hürland M., Kuhlgatz D.A., Kuhlgatz C., Osmers J.H., Jung M., Schulze M.  2023. The use of machine learning methods to predict sperm quality in Holstein bulls. Theriogenology. 197:16–25. doi:10.1016/j.theriogenology.2022.11.032 [DOI] [PubMed] [Google Scholar]
  22. Laplacette A.L., Giordano J.O.  2025. Predicting first service pregnancy outcome in lactating dairy cows with machine learning algorithms trained on behavioral, performance, and cow features data. J. Dairy Sci. 108 (E-Suppl. 1). [Google Scholar]
  23. Lou W., Bonfatti V., Bovenhuis H., Shi R., van der Linden A., Mulder H.A., Liu L., Wang Y., Ducro B.  2024. Prediction of likelihood of conception in dairy cows using milk mid-infrared spectra collected before the first insemination and machine learning algorithms. J. Dairy Sci. 107(11):9415–9425. 10.3168/jds.2023-24621 [DOI] [PubMed] [Google Scholar]
  24. Marín-Urías L.F., García-Ramírez P.J., Domínguez-Mancera B., Hernández-Beltrán A., Vásquez-Santacruz J.A., Cervantes-Acosta P., Barrientos-Morales M., Portillo-Vélez R. d J.  2023. Bull breeding soundness assessment using artificial neural network-based predictive models. Agriculture. 14(1):67. doi.org/10.3390/agriculture14010067 [Google Scholar]
  25. Marques T.C., Marques L.R., Fernandes P.B., de Lima F.S., do Prado Paim T., Leão K.M.  2024. Machine learning to predict pregnancy in dairy cows: an approach integrating automated activity monitoring and on-farm data. Animals. 14(11):1567. 10.3390/ani14111567 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Mayo L.M., Silvia W.J., Ray D.L., Jones B.W., Stone A.E., Tsai I.C., Clark J.D., Bewley J.M., Heersche G.  2019. Automated estrous detection using multiple commercial precision dairy monitoring technologies in synchronized dairy cows. J. Dairy Sci. 102(3):2645–2656. doi: 10.3168/jds.2018-14738 [DOI] [PubMed] [Google Scholar]
  27. McDonagh J., Tzimiropoulos G., Slinger K.R., Huggett Z.J., Down P.M., Bell M.J.  2021. Detecting dairy cow behavior using vision technology. Agriculture. 11(7):675. 10.3390/agriculture11070675 [DOI] [Google Scholar]
  28. Mendonça L.G.D., Wronka K., Fridkovski G., Bilby T., Stevenson J.  2025. Reproductive efficiency of Low-Fertility dairy cows submitted to Presynch-Ovsynch or Double-Ovsynch scheme for first service. Animals. 15(4):536. 10.3390/ani15040536 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Perez M.M., Cabrera E., Rial C., You Y., Wang Y., Weinberger K., Nydam D.V., Giordano J.O.  2025. Screening and selection of a machine learning algorithm for development of a model to select cows for clinical examination using data from automated health monitoring technologies and other predictors of cow health. J. Dairy Sci. 108(12):13720–13738. 10.3168/jds.2025-26511 [DOI] [PubMed] [Google Scholar]
  30. Raes A., Babin D., Pascottini O.B., Opsomer G., Van Soom A., Smits K.  2025. Artificial intelligence outperforms humans in morphology-based oocyte selection in cattle. Sci. Rep. 15(1):21829. 10.1038/s41598-025-09019-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Ratkovic M.  2023. Relaxing assumptions, improving inference: Integrating machine learning and the linear regression. Am. Polit. Sci. Rev. 117(3):1053–1069. 10.1017/S0003055422001022 [DOI] [Google Scholar]
  32. Rial C., Laplacette A., Giordano J.O.  2022. Effect of a targeted reproductive management program designed to prioritize insemination at detected estrus and optimize time to insemination on the reproductive performance of lactating dairy cows. J. Dairy Sci. 105(10):8411–8425. doi: 10.3168/jds.2022-22082 [DOI] [PubMed] [Google Scholar]
  33. Samuel A.L.  1959. Some studies in machine learning using the game of checkers. IBM J. Res. Dev. 3(3):210–229. doi: 10.1147/rd.33.0210 [DOI] [Google Scholar]
  34. Sarker I.H.  2021. Machine learning: algorithms, real-world applications and research directions. SN Comput. Sci. 2(3):160. doi: 10.1007/s42979-021-00592-x [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Schilkowsky E.M., Granados G.E., Sitko E.M., Masello M., Perez M.M., Giordano J.O.  2021. Evaluation and characterization of estrus alerts and behavioral parameters generated by an ear-attached accelerometer-based system for automated detection of estrus. J. Dairy Sci. 104(5):6222–6237. doi: 10.3168/jds.2020-19667 [DOI] [PubMed] [Google Scholar]
  36. Shahinfar S., Page D., Guenther J., Cabrera V., Fricke P., Weigel K.  2014. Prediction of insemination outcomes in Holstein dairy cattle using alternative machine learning algorithms. J. Dairy Sci. 97(2):731–742. doi: 10.3168/jds.2013-6693 [DOI] [PubMed] [Google Scholar]
  37. Shahriar M.S., Smith D., Rahman A., Freeman M., Hills J., Rawnsley R., Henry D., Bishop-Hurley G.  2016. Detecting heat events in dairy cows using accelerometers and unsupervised learning. Comput. Electron. Agric. 128 (Oct 2016):20–26. doi: 10.1016/j.compag.2016.08.009 [DOI] [Google Scholar]
  38. Sitko E.M., Perez M.M., Granados G.E., Masello M., Sosa Hernandez F., Cabrera E.M., Schilkowsky E.M., Di Croce F.A., McNeel A.K., Weigel D.J.  et al.  2023. Effect of reproductive management programs that prioritized artificial insemination at detected estrus or timed artificial insemination on the reproductive performance of primiparous Holstein cows of different genetic merit for fertility. J. Dairy Sci. 106(9):6495–6514. doi: 10.3168/jds.2022-22673 [DOI] [PubMed] [Google Scholar]
  39. Tsai D.-M., Huang C.-Y.  2014. A motion and image analysis method for automatic detection of estrus and mating behavior in cattle. Comput. Electron. Agric. 104:25–31. doi: 10.1016/j.compag.2014.03.003 [DOI] [Google Scholar]
  40. Urli S., Corte Pause F., Dreossi T., Crociati M., Stradaioli G.  2025. Evaluation of an artificial intelligence system for bull sperm morphology evaluation. Theriogenology. 245:117504. doi: 10.1016/j.theriogenology.2025.117504 [DOI] [PubMed] [Google Scholar]
  41. Wang J., Bell M., Liu X., Liu G.  2020. Machine-learning techniques can enhance dairy cow estrus detection using location and acceleration data. Animals. 10(7):1160. 10.3390/ani10071160 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Wang Z., Hua Z., Wen Y., Zhang S., Xu X., Song H.  2024. E-YOLO: Recognition of estrus cow based on improved YOLOv8n model. Expert Syst. Appl. 238 (Part E):122212. doi.org/10.1016/j.eswa.2023.122212. [Google Scholar]
  43. You J., Ellis J.L., Adams S., Sahar M., Jacobs M., Tulpan D.  2023. Comparison of imputation methods for missing production data of dairy cattle. Animal. 17  Suppl 5:100921. doi: 10.1016/j.animal.2023.100921 [DOI] [PubMed] [Google Scholar]

Articles from Animal Frontiers: The Review Magazine of Animal Agriculture are provided here courtesy of Oxford University Press

RESOURCES