CoChat
Sign Up
Cited by 505Wildlife Ecology and ConservationAnimal Behavior and ReproductionData Mining and Machine Learning Applications

Applications of machine learning in animal behaviour studies

Open access
Valletta, John Joseph, Torney, Colin, Kings, Michael, Thornton, Alex, Madden, Joah·2017·Animal Behaviour
DOI 10.1016/j.anbehav.2016.12.005
Open this paper in CoChat — free accountOpen PDF

Abstract

In many areas of animal behaviour research, improvements in our ability to collect large and detailed data sets are outstripping our ability to analyse them. These diverse, complex and often high-dimensional data sets exhibit nonlinear dependencies and unknown interactions across multiple variables, and may fail to conform to the assumptions of many classical statistical methods. The field of machine learning provides methodologies that are ideally suited to the task of extracting knowledge from these data. In this review, we aim to introduce animal behaviourists unfamiliar with machine learning (ML) to the promise of these techniques for the analysis of complex behavioural data. We start by describing the rationale behind ML and review a number of animal behaviour studies where ML has been successfully deployed. The ML framework is then introduced by presenting several unsupervised and supervised learning methods. Following this overview, we illustrate key ML approaches by developing data analytical pipelines for three different case studies that exemplify the types of behavioural and ecological questions ML can address. The first uses a large number of spectral and morphological characteristics that describe the appearance of pheasant, Phasianus colchicus, eggs to assign them to putative clutches. The second takes a continuous data stream of feeder visits from PIT (passive integrated transponder)-tagged jackdaws, Corvus monedula, and extracts foraging events from it, which permits the construction of social networks. Our final example uses aerial images to train a classifier that detects the presence of wildebeest, Connochaetes taurinus, to count individuals in a population. With the advent of cheaper sensing and tracking technologies an unprecedented amount of data on animal behaviour is becoming available. We believe that ML will play a central role in translating these data into scientific knowledge and become a useful addition to the animal behaviourist's analytical toolkit.

Verified 2026-10-09 · 2 references not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

This review introduces machine learning to animal behaviourists and illustrates unsupervised and supervised workflows through three case studies: assigning pheasant eggs to clutches, extracting jackdaw foraging events to build social networks, and detecting wildebeest in aerial images. It argues ML is well suited to complex, high-dimensional behavioural data but is not a silver bullet and requires careful validation.

Key findings

  • Pheasant eggs: 549 eggs from 30 females in 10 pens were described by 21 morphological and spectral measures; PCA retained four components explaining 96% of variance, and k-means/hierarchical clustering with silhouette width indicated k = 3 clutches.
  • Jackdaws: around 1000 PIT-tagged individuals across three colonies were studied; an illustrative 4-week data set from 141 individuals at one feeder was modelled with a Gaussian mixture model and BIC, and the resulting network confirmed associations between mated adults, parents and offspring, and unrelated individuals.
  • Wildebeest: 250 aerial images of 64 × 64 pixels (50 each for wildebeest, zebra, trees, rocks and grass) were classified using HOG features and random forests; 175 images were used for training and 75 for testing, and from over 1000 features only a couple of hundred were needed, though two wildebeest in the test set were misclassified as trees.
  • The review states ML is hypothesis-free and prediction-focused, can handle nonlinearities and interactions, but can learn patterns different from human cues and is vulnerable to biased training data and mistaking correlation for causation.

Methods

  • Narrative review plus three worked case studies, with documented code at https://github.com/jjvalletta/ML_Animal_Behaviour; methods covered include PCA, k-means, hierarchical clustering, Gaussian mixture models, decision trees, random forests, cross-validation and feature selection.
  • Case-study data: pheasant eggs (30 females, 10 pens, 549 eggs, 21 morphological/spectral measures, reflectance spectra 300–700 nm); jackdaw PIT-tag feeder visits (around 1000 birds across three colonies; illustrative 4 weeks from 141 individuals at one feeder); wildebeest aerial photos (250 images, 64 × 64 pixels, five classes).
  • Main analyses: PCA and clustering for pheasant eggs; GMM with BIC and adjacency-matrix construction for jackdaw social networks; HOG feature extraction, t-SNE visualization, random forests with variable importance and a confusion matrix for wildebeest detection.

Limitations

  • The paper states ML is “no silver bullet”: models may use cues unlike human cues (e.g. wildebeest misclassified as trees), biased training data can cause erratic behaviour (the tank/no-tank anecdote), observational data can mistake correlation for causation, and reproducibility over independent data sets is the most rigorous validation.
  • The paper states that determining the number of clusters k is a fundamental yet unsolved problem, is mathematically subjective, and the final decision rests with the practitioner.
  • Evident from the text: the case studies are illustrative rather than fully validated—pheasant maternity validation by protein fingerprinting is proposed but not performed; the jackdaw analysis uses only 4 weeks, one feeder and 141 individuals; and the wildebeest classifier uses five classes for illustration although automated counting would require only two.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1016/j.anbehav.2016.12.005

Cites

97 of 99 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 49 cited sources checked · 0 supported · 0 flagged

  • Findingreasoning

    PC3 is a measure of egg size.

    “Thus, PC1 is a measure of eggshell brightness, PC2 is a measure of eggshell greenness, PC3 is a measure of egg size and PC4 is a measure of egg shape.”
    Quote in the paperNot a substantial claim
  • Findingown data

    The jackdaw social network generated confirms the existence of foraging associations between mated adults and between parents and offspring.

    “The jackdaw social network we generated (see Fig. 4e) confirms the existence of foraging associations between mated adults and between parents and offspring.”
    Quote in the paperSubstantial claim
  • Findingown data

    The jackdaw intervisit time data are dominated by short intervisit periods corresponding to foraging events, interleaved with long periods of inactivity.

    “As depicted in Fig. 4b, the data are dominated by short intervisit periods corresponding to foraging events, interleaved with long periods of inactivity.”
    Quote in the paperSubstantial claim
  • Findingown data

    The first four principal components explained 96% of the variance in the pheasant egg data.

    “Fig. 3c depicts the PCA outcome, where the first four components explained 96% of the variance in the data and were kept as features for subsequent analysis.”
    • Also stated in: § CONCLUDING REMARKS · plain
      “The first four components explained 96% of the variance in the data and were kept as features for subsequent analysis.”
    Quote in the paperNot a substantial claim
  • Findingown data

    From an initial pool of over 1000 features, only a couple of hundred predictors were needed to achieve good predictive performance on the training data set.

    “From an initial pool of over 1000 features only a couple of hundred predictors were needed to achieve good predictive performance on the training data set.”
    Quote in the paperNot a substantial claim
  • May combine two claims. 29
  • Classified as probably not a substantial claim. 39

  • May combine two claims.

    Findingown data

    When HOGs features were projected onto the first two PCs, they showed promising grouping by class, although it was still somewhat concealed.

    “When projected onto the first two PCs, the HOGs features showed promising grouping by class, although it was still somewhat concealed (see Appendix Fig. A5b).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The jackdaw social network reveals strong associations between unrelated individuals and shows how these associations relate to the spatial proximity between nestboxes.

    “It also reveals strong associations between unrelated individuals and shows how these associations relate to the spatial proximity between nestboxes.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The confusion matrix in Fig. 5e shows how two wildebeests in the test data were mistakenly labelled as trees.

    “For example, the confusion matrix in Fig. 5e shows how two wildebeests in the test data were mistakenly labelled as trees.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    Visual inspection of pheasant eggs suggested that within-individual variation in egg morphology may be low while inter-individual variation is high.

    “Visual inspection of eggs suggested that withinindividual variation in egg morphology may be low while interindividual variation is high.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    For statistical models, the focus is on inference: estimating the parameters of the statistical model that most likely gave rise to the observed data and providing uncertainty bounds for these estimates.

    “The focus is on inference; estimating the parameters of the statistical model that most likely gave rise to the observed data, and providing uncertainty bounds for these estimates.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Breiman, 2001b

      Not checked in the free run.

    • cites Friedman, 2001 source

      Not checked in the free run.

    • cites Zoubin Ghahramani, 2015

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Statistical models start with an assumption about the underlying data distribution, such as Gaussian or Poisson.

    “Statistical models start with an assumption about the underlying data distribution (e.g.Gaussian, Poisson).”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Breiman, 2001b

      Not checked in the free run.

    • cites Friedman, 2001 source

      Not checked in the free run.

    • cites Zoubin Ghahramani, 2015

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisereasoning

    Machine learning allows important questions to be answered across a range of animal behaviour topics, including movement ecology, social structure, collective behaviour, communication and welfare.

    “This allows us to answer important questions across a range of topics, including movement ecology, social structure, collective behaviour, communication and welfare.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: In many areas of animal behaviour research, improvements in the ability to collect large and detailed data sets are outstripping the ability to analyse them.
    • Rests on an assumption: Animal behaviour researchers are in a position where automatically collecting detailed data sets is becoming commonplace, but extracting knowledge from them is a daunting task, mainly due to the lack of accessible analytical tools.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The field of machine learning provides methodologies that are ideally suited to extracting knowledge from complex behavioural data.

    “The field of machine learning provides methodologies that are ideally suited to the task of extracting knowledge from these data.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: In many areas of animal behaviour research, improvements in the ability to collect large and detailed data sets are outstripping the ability to analyse them.
    • Rests on an assumption: These diverse, complex and often high-dimensional data sets exhibit nonlinear dependencies and unknown interactions across multiple variables, and may fail to conform to the assumptions of many classical statistical methods.
    • Rests on an assumption: Animal behaviour researchers are in a position where automatically collecting detailed data sets is becoming commonplace, but extracting knowledge from them is a daunting task, mainly due to the lack of accessible analytical tools.

  • May combine two claims.

    Premisecitation

    Statistical modelling and machine learning both seek to build a mathematical description, a model, of the data and the underlying mechanism it represents, and therefore substantially overlap.

    “Both statistical modelling and ML seek to build a mathematical description, a model, of the data and the underlying mechanism it represents; thus inevitably there is substantial overlap between the two”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Breiman, 2001b

      Not checked in the free run.

    • cites Friedman, 2001 source

      Not checked in the free run.

    • cites Zoubin Ghahramani, 2015

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    Machine learning approaches can address otherwise intractable tasks in animal behaviour, such as classifying species, individuals, vocalizations or behaviours within complex data sets.

    “In animal behaviour, ML approaches can address otherwise intractable tasks, such as classifying species, individuals, vocalizations or behaviours within complex data sets.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: In many areas of animal behaviour research, improvements in the ability to collect large and detailed data sets are outstripping the ability to analyse them.
    • Rests on an assumption: Animal behaviour researchers are in a position where automatically collecting detailed data sets is becoming commonplace, but extracting knowledge from them is a daunting task, mainly due to the lack of accessible analytical tools.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    For machine learning, the focus is typically on prediction; without necessarily assuming a functional distribution for the data, a model that achieves optimal predictive performance is identified.

    “For ML, the focus is typically on prediction; without necessarily assuming a functional distribution for the data, a model that achieves optimal predictive performance is identified.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Breiman, 2001b

      Not checked in the free run.

    • cites Friedman, 2001 source

      Not checked in the free run.

    • cites Zoubin Ghahramani, 2015

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Machine learning was used in a comparative assessment of welfare across multiple laboratory populations of mice, permitting a wide range of potential explanatory factors with diverse distributions to be considered simultaneously as well as their interactions.

    “ML was used in a comparative assessment of welfare across multiple laboratory populations of mice (Chesler et al., 2002) permitting a wide range of potential explanatory factors, each with diverse distribution, to be considered simultaneously as well as the interactions between them.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Chesler et al., 2002 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Individual pixel intensities are not explicitly related to the image's class, suggesting the need for hand-crafted features.

    “In this case, the individual pixel intensities are not explicitly related to the image's class, suggesting the need for hand-crafted features that describe the object in each image.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    Out-of-bag data are used to compute an estimate of the importance of every predictor, which can subsequently be used for feature selection.

    “The OOB data are also used to compute an estimate of the importance of every predictor, which can be subsequently used for feature selection.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Using such a classifier may soon enable the fully automated detection and identification of animals from aerial count images.

    “Using such a classifier may soon enable the fully automated detection and identification of animals from aerial count images.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Machine learning can relieve this burden by detecting wildebeest in an image and therefore automating this aspect of the count.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Random forests is an ensemble method developed to mitigate the problem of overfitting in decision trees.

    “Random forests (Breiman, 2001a;Cutler et al., 2007) is an ensemble method developed to mitigate the problem of overfitting in decision trees.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Breiman, 2001a

      Not checked in the free run.

    • cites Cutler et al., 2007 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Traditional statistical modelling puts forward a hypothesis and accepts or rejects it depending on consistency with measured observations, whereas machine learning methods learn the hypothesis directly from the training data set.

    “While in traditional statistical modelling a hypothesis (model) is put forward and is then accepted/rejected depending on how consistent it is with the measured observations, ML methods learn this hypothesis directly from the training data set.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Breiman, 2001b

      Not checked in the free run.

    • cites Friedman, 2001 source

      Not checked in the free run.

    • cites Zoubin Ghahramani, 2015

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisereasoning

    Machine learning algorithms can deal with nonlinearities and interactions among variables because the models are flexible enough to fit the data.

    “ML algorithms can deal with nonlinearities and interactions among variables because the models are flexible enough to fit the data (as opposed to rigid linear regression models, for example).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    Random forests can handle thousands of mixed categorical and continuous predictors and are robust to outliers, but the interpretability of plain decision trees is lost.

    “Random forests can handle thousands of mixed categorical and continuous predictors and are robust to outliers; however, the interpretability of plain decision trees is lost.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisereasoning

    The flexibility of machine learning algorithms needs to be constrained to avoid fitting noise, or overfitting.

    “However, this flexibility needs to be constrained to avoid fitting noise (overfitting).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    Machine learning can address different objectives in numerous distinct fields of animal behaviour and is becoming a staple approach for novel methods.

    “It is clear that ML can address different objectives in numerous distinct fields of animal behaviour and is thus becoming a staple approach for novel methods (Dankert, Wang, Hoopfer, Anderson, & Perona, 2009;Kabra et al., 2013).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Dankert et al., 2009 source

      Not checked in the free run.

    • cites Kabra et al., 2013 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Finer scale discriminations are possible at both the individual level and the bird song elements level.

    “Finer scale discriminations are possible at both the individual level (Cheng, Xie, Lin, & Ji, 2012) and the bird song elements level (Ranjard & Ross, 2008).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Cheng et al., 2012 source

      Not checked in the free run.

    • cites Ranjard & Ross, 2008

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Co-occurrences in time can be determined by machine learning clustering methods, with individuals in the same foraging event considered to have a social affiliation.

    “Co-occurrences in time could be determined by ML clustering methods with individuals in the same foraging event (cluster) considered to have a social affiliation (Psorakis et al., 2015).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Psorakis et al., 2015 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Female pheasants may nest communally or lay eggs in multiple nests.

    “Female pheasants may nest communally or lay eggs in multiple nests (Yom-Tov, 1980).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Yom-Tov, 1980 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    k-means clustering is popular because it is intuitive and computationally inexpensive.

    “k-means clustering is popular because it is intuitive and computationally inexpensive.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Machine learning is used to automate the classification of behaviours or activities and tracking movement trajectories from GPS, accelerometer and/or video data.

    “ML is used to automate the classification of behaviours/activities (Kabra et al., 2013) and tracking movement trajectories (Dell et al., 2014).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Kabra et al., 2013 source

      Not checked in the free run.

    • cites Dell et al., 2014 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The hypothesis-free approach of machine learning makes it an attractive choice for dealing with complex data sets.

    “It is this hypothesis-free approach that makes ML an attractive choice for dealing with complex data sets.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    Decision trees are prone to overfitting and their predictive performance can be poor compared to other algorithms.

    “However, they are prone to overfitting (small changes in the training data set may lead to significantly different trees) and their predictive performance can be poor compared to other algorithms.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationcitation

    If achieved, this will have major implications for conservation as currently the process of detecting and identifying animals within images is the main bottleneck for estimating population sizes from aerial surveys.

    “If achieved, this will have major implications for conservation as currently the process of detecting and identifying animals within images is the main bottleneck for estimating population sizes from aerial surveys (Torney et al., 2016).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Torney et al., 2016 source

      Not checked in the free run.

    • Rests on an assumption: Currently counting wildebeest from aerial photos is time consuming, as thousands of aerial photos need to be counted manually.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    Gaussian mixture models assume that data are generated from a mixture of multivariate Gaussians and lack a global optimum solution.

    “GMMs, however, assume that data are generated from a mixture of multivariate Gaussians and lack a global optimum solution.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Premisereasoning

    Determining the number of distinct clusters k in a data set is a fundamental yet unsolved problem.

    “Determining the number of distinct clusters k in a data set is a fundamental yet unsolved problem.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Machine learning clustering methods for social affiliation appear to be robust and capture real-life pair bonds well.

    “Such methods appear to be robust and capture real-life pair bonds well (Psorakis, Roberts, Rezek, & Sheldon, 2012).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Psorakis et al., 2012 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    Decision trees are readily interpretable, directly used to generate rules, and computationally inexpensive to train, evaluate and store.

    “Decision trees are readily interpretable, directly used to generate rules, and computationally inexpensive to train, evaluate and store.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisereasoning

    Although collected data may seem high dimensional, the structure of the data can be represented by fewer variables.

    “The rationale behind dimensionality reduction is simple; although the collected data may seem high dimensional, the structure of the data can be represented by fewer variables.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Using unsupervised learning methods, the authors were able to reduce the dimensionality of the pheasant egg problem to a few biologically meaningful features and subsequently assign eggs to females.

    “Using unsupervised learning methods, we were able to reduce the dimensionality of the problem to a few biologically meaningful features, and subsequently assign eggs to females.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Without direct observation of laying or destructive genetic sampling from eggs, it is not possible to assign eggs to females.

  • Classified as probably not a substantial claim.

    Premisecitation

    Based on Psorakis et al. (2012), the workflow converts a long series of time stamps into a social network.

    “Based on the work of Psorakis et al. (2012), Fig. 4a summarizes the workflow used to convert a long series of time stamps into a social network.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Psorakis et al. (2012) source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    Decision trees can handle both categorical and continuous data and are fairly robust to outliers.

    “They can handle both categorical and continuous data, and are fairly robust to outliers.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Premisenone given

    k-means clustering is only applicable to continuous data where a mean is defined.

    “However, it is only applicable to continuous data where a mean is defined.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisenone given

    Feature extraction is a key step in machine learning, and finding representations directly relevant to the task will almost always result in better predictive accuracy than employing more complex models.

    “Feature extraction is a key step in ML; finding representations that are directly relevant to the task at hand (e.g.discriminating between two classes) will almost always result in better predictive accuracy than employing more complex models.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisenone given

    k-means clustering has no guarantee of a global optimum solution, so the user is advised to start the algorithm at multiple distinct centroids.

    “There is also no guarantee of a global optimum solution; thus, the user is advised to start the algorithm at multiple distinct centroids.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Machine learning can tackle a wide range of tasks, including classifying observations into predefined sets, clustering data into groups that share an underlying process, and regressing an outcome against multiple factors and elucidating their contributory effect.

    “ML can tackle a wide range of tasks, including classifying observations into predefined sets (Kabra, Robie, Rivera-Alba, Branson, & Branson, 2013), clustering data into groups that share an underlying process (Zhang, O'Reilly, Perry, Taylor, & Dennis, 2015) and regressing an outcome of interest against multiple factors and elucidating their contributory effect (Chesler, Wilson, Lariviere, Rodrig”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Kabra et al., 2013 source

      Not checked in the free run.

    • cites Zhang et al., 2015 source

      Not checked in the free run.

    • cites Chesler et al., 2002 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Machine learning-derived knowledge can be used to infer individual decision rules in collective motion and to compute activity budgets for individuals without continuous human observation or time-consuming video analysis.

    “This knowledge can then be used to infer individual decision rules in collective motion (Katz, Tunstrom, Ioannou, Huepe, & Couzin, 2011;Nagy, Kos, Biro, & Vicsek, 2010) and to compute activity budgets for individuals without the need for continuous human observation or time-consuming video analysis.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Katz et al., 2011 source

      Not checked in the free run.

    • cites Nagy et al., 2010 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    There are cases where a mix of labelled and unlabelled data are available, which would require the use of semisupervised learning.

    “There are also cases where a mix of labelled and unlabelled data are available, which would require the use of a different type of ML methods: semisupervised learning (Chapelle, Sch€ olkopf, & Zien, 2006).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Chapelle, Schölkopf, & Zien, 2006

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationown data

    PC3 is a measure of egg size in the pheasant egg data.

    “PC1 is a measure of eggshell brightness, PC2 is a measure of eggshell greenness, PC3 is a measure of egg size and PC4 is a measure of egg shape.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationreasoning

    Unsupervised learning methods are routinely used with labelled data during the data exploration stage.

    “However, as we have seen earlier, unsupervised learning methods are routinely used with labelled data during the data exploration stage.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationown data

    PC4 is a measure of egg shape in the pheasant egg data.

    “PC1 is a measure of eggshell brightness, PC2 is a measure of eggshell greenness, PC3 is a measure of egg size and PC4 is a measure of egg shape.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Automatically extracting predictive features from 'raw' data is the central theme of Deep Learning.

    “To this end, automatically extracting predictive features from 'raw' data is the central theme of a new set of methods, in an active area of research, called Deep Learning (LeCun, Bengio, & Hinton, 2015).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites LeCun, Bengio, & Hinton, 2015 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    High-profile failures highlight how even experts in the field of ML can fall victim to common pitfalls in data-driven analytics.

    “High-profile failures highlight how even experts in the field of ML can fall victim of common pitfalls in data-driven analytics.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    The type of data (labelled or unlabelled) and the particular task at hand ultimately dictate the category of algorithms to choose from.

    “The type of data (labelled or unlabelled) and the particular task at hand ultimately dictate the category of algorithms to choose from (see Table 1).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Deep Learning automatically discovers features amenable for prediction by recursively applying simple but nonlinear transformations to the data.

    “Instead of using hand-crafted features (as we did in the wildebeest identification problem), Deep Learning automatically discovers features amenable for prediction by recursively applying simple but nonlinear transformations to the data.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites LeCun, Bengio, & Hinton, 2015 source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Associations between an outcome and a set of inputs could have occurred by chance, owing to a confounding factor or a biased data set.

    “Associations between an outcome and a set of inputs could have occurred by chance, owing to a confounding factor or a biased data set.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites MacKinnon, Krull, & Lockwood, 2000 source

      Not checked in the free run.

    • cites Zuur, Ieno, & Elphick, 2010

      Not checked in the free run.

    • cites A. Liu & Ziebart, 2014

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationnone given

    Reproducibility of a predictive model over multiple independent data sets over time is ultimately the most rigorous form of validation.

    “Reproducibility of a predictive model over multiple independent data sets over time is ultimately the most rigorous form of validation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Semisupervised learning methods make use of unlabelled data to improve the performance of predictive models identified from labelled observations.

    “These methods make use of unlabelled data to improve the performance of predictive models identified from labelled observations.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Chapelle, Schölkopf, & Zien, 2006

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    There are no hard and fast rules to choosing an algorithm, and the performance of several state-of-the-art supervised learning algorithms are similar.

    “However, there are no hard and fast rules to this process and it has been shown that the performance of several state-of-the-art supervised learning algorithms are similar (Fern andez-Delgado, Cernadas, Barro, & Amorim, 2014).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Fernandez-Delgado, Cernadas, Barro, & Amorim, 2014

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationown data

    PC1 is a measure of eggshell brightness in the pheasant egg data.

    “PC1 is a measure of eggshell brightness, PC2 is a measure of eggshell greenness, PC3 is a measure of egg size and PC4 is a measure of egg shape.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationown data

    PC2 is a measure of eggshell greenness in the pheasant egg data.

    “PC1 is a measure of eggshell brightness, PC2 is a measure of eggshell greenness, PC3 is a measure of egg size and PC4 is a measure of egg shape.”
    Quote in the paperNot a substantial claim
56 of 56

  • The paper relies on this without support.

    In many areas of animal behaviour research, improvements in the ability to collect large and detailed data sets are outstripping the ability to analyse them.

    “In many areas of animal behaviour research, improvements in our ability to collect large and detailed data sets are outstripping our ability to analyse them.”

  • The paper relies on this without support.

    These diverse, complex and often high-dimensional data sets exhibit nonlinear dependencies and unknown interactions across multiple variables, and may fail to conform to the assumptions of many classical statistical methods.

    “These diverse, complex and often highdimensional data sets exhibit nonlinear dependencies and unknown interactions across multiple variables, and may fail to conform to the assumptions of many classical statistical methods.”

  • The paper relies on this without support.

    Animal behaviour researchers are in a position where automatically collecting detailed data sets is becoming commonplace, but extracting knowledge from them is a daunting task, mainly due to the lack of accessible analytical tools.

    “Animal behaviour researchers are thus in a position where automatically collecting detailed data sets is becoming commonplace, but extracting knowledge from them is a daunting task, mainly due to the lack of accessible analytical tools.”

  • The paper relies on this without support.

    Without direct observation of laying or destructive genetic sampling from eggs, it is not possible to assign eggs to females.

    “Without direct observation of laying or destructive genetic sampling from eggs, it is not possible to assign eggs to females.”

  • The paper relies on this without support.

    Uncovering social structures in jackdaws allows us to infer individualized social relationships and determine how novel information spreads through social groups.

    “Uncovering social structures in jackdaws allows us to infer individualized social relationships and determine how novel information spreads through social groups.”

  • The paper relies on this without support.

    It is unclear how one specifies the length of foraging events without fixing an arbitrary time window.

    “It is unclear, however, how one specifies the length of these foraging events without fixing an arbitrary time window.”

  • The paper relies on this without support.

    Currently counting wildebeest from aerial photos is time consuming, as thousands of aerial photos need to be counted manually.

    “Currently this process is time consuming, as thousands of aerial photos need to be counted manually.”

  • The paper relies on this without support.

    Machine learning can relieve this burden by detecting wildebeest in an image and therefore automating this aspect of the count.

    “Machine learning can relieve this burden by detecting wildebeest in an image and therefore automating this aspect of the count.”

  • The paper relies on this without support.

    Research on jackdaw social structure has largely been confined to captive studies and there is limited information on patterns of association in natural populations.

    “However, research has largely been confined to captive studies and there is limited information on patterns of association in natural populations.”

  • The paper relies on this without support.

    HOGs features are hard to interpret and are likely to exhibit nonlinear interactions among them.

    “HOGs features are hard to interpret and are likely to exhibit nonlinear interactions among them (a stark contrast to the pheasant eggs data where PCA was justified by the linear relationships between variables of the same theme).”

  • The paper relies on this without support.

    Biased training data sets cause erratic behaviour of the predictive model.

    “For example, biased training data sets cause erratic behaviour of the predictive model.”

  • The paper relies on this without support.

    Features that are directly relevant to the outcome being predicted tend to make the predictive performance insensitive to the choice of algorithm.

    “Features that are directly relevant to the outcome being predicted tend to make the predictive performance insensitive to the choice of algorithm.”

  • The paper relies on this without support.

    ML offers a hypothesis-free approach to model complex data sets where the type of relationship between measured variables is unknown.

    “ML offers a hypothesis-free approach to model complex data sets where the type of relationship between measured variables is unknown.”

  • The paper relies on this without support.

    ML algorithms mine for patterns in the data that are best at predicting an outcome.

    “ML algorithms mine for patterns in the data that are best at predicting an outcome.”

  • The paper relies on this without support.

    Typically, ML algorithms mine for patterns in observational data, rather than experimental data, where correlation can be mistaken for causation.

    “Typically, ML algorithms mine for patterns in observational data, rather than experimental data, where correlation can be mistaken for causation.”
15 of 15

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Opposite objectives stated for separation between clusters

    Minor

    The clustering section says to maximize similarity between clusters, while Figure A1 says to maximize the distance between clusters. Those are opposite objectives.

  • Opposite objectives stated for similarity within clusters

    Minor

    The clustering section says observations within a group should be similar, but Figure A1 says to minimize within-group similarity.

2 of 2

Cites

Flagged first, then by citations

5 of the 104 references couldn’t be matched to a known paper.

Research works printed but not linked: 3

  1. Bishop, 2006, Vol. 4
  2. Broom, 1993, Vol. 30
  3. Hastie, 2009, Vol. 1

Cited by

Sorted by citations

Related

Authors
Valletta, J. J., Torney, C., Kings, M., Thornton, A., & Madden, J.
Affiliations
University of Exeter (GB)
Corresponding author
Valletta, John Joseph
ORCID
Venue
Animal Behaviour
Publisher
Elsevier BV
Published
2017-02
Volume
124
Pages
203-220
ISSN
0003-3472 (ISSN-L)
DOI
10.1016/j.anbehav.2016.12.005
OpenAlex
W2581974582
Cited by
505
References
104
Citations by year
2026: 29 · 2025: 57 · 2024: 68 · 2023: 70 · 2022: 84 · 2021: 69 · 2020: 47 · 2019: 48 · 2018: 27 · 2017: 5
FWCI
26.26
Type
journal article
Language
en
Open access
Licence
Locations
  • Animal Behaviour (publishedVersion, cc-by)
  • ENLIGHTEN (Jurnal Bimbingan dan Konseling Islam) (acceptedVersion)
  • Open Research Exeter (University of Exeter) (acceptedVersion, cc-by)
  • Figshare (submittedVersion)
Funders
  • MRC (MR/M003906/1)
  • BBSRC David Phillips Fellowship (BB/H021817/1)
  • BBSRC SWDTP studentship (630051486)
  • ERC consolidator award (616474)
Topics
  • Physical Sciences › Environmental Science › Ecology › Wildlife Ecology and Conservation
  • Life Sciences › Agricultural and Biological Sciences › Ecology, Evolution, Behavior and Systematics › Animal Behavior and Reproduction
  • Physical Sciences › Computer Science › Information Systems › Data Mining and Machine Learning Applications
Keywords
machine learning, behavioural data analysis, unsupervised learning, supervised learning, jackdaw, social networks, aerial imagery, population counting, passive integrated transponders

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review