Creating a behavioural classification module for acceleration data: using a captive surrogate for difficult to observe species

Opportunities to monitor animal welfare using the five freedoms with precision livestock management on rangelands

Advances in technology have led to precision livestock management, a developing research field. Precision livestock management has potential to improve sustainable meat production through continuous, real-time tracking which can help livestock managers remotely monitor and enhance animal welfare in extensive rangeland systems. The combination of global positioning systems (GPS) and accessible data transmission gives livestock managers the ability to locate animals in arduous weather, track animal patterns throughout the grazing season, and improve handling practices. Accelerometers fitted to ear tags or collars have the potential to identify behavioral changes through variation in the intensity of movement that can occur during grazing, the onset of disease, parturition or responses to other environmental and management stressors. The ability to remotely detect disease, parturition, or effects of stress, combined with appropriate algorithms and data analysis, can be used to notify livestock managers and expedite response times to bolster animal welfare and productivity. The “Five Freedoms” were developed to help guide the evaluation and impact of management practices on animal welfare. These freedoms and welfare concerns differ between intensive (i.e., feed lot) and extensive (i.e., rangeland) systems. The provisions of the Five Freedoms can be used as a conceptual framework to demonstrate how precision livestock management can be used to improve the welfare of livestock grazing on extensive rangeland systems.

What Are Sheep Doing? Tri-Axial Accelerometer Sensor Data Identify the Diel Activity Pattern of Ewe Lambs on Pasture

Monitoring activity patterns of animals offers the opportunity to assess individual health and welfare in support of precision livestock farming. The purpose of this study was to use a triaxial accelerometer sensor to determine the diel activity of sheep on pasture. Six Perendale ewe lambs, each fitted with a neck collar mounting a triaxial accelerometer, were filmed during targeted periods of sheep activities: grazing, lying, walking, and standing. The corresponding acceleration data were fitted using a Random Forest algorithm to classify activity (=classifier). This classifier was then applied to accelerometer data from an additional 10 ewe lambs to determine their activity budgets. Each of these was fitted with a neck collar mounting an accelerometer as well as two additional accelerometers placed on a head halter and a body harness over the shoulders of the animal. These were monitored continuously for three days. A classification accuracy of 89.6% was achieved for the grazing, walking and resting activities (i.e., a new class combining lying and standing activity). Triaxial accelerometer data showed that sheep spent 64% (95% CI 55% to 74%) of daylight time grazing, with grazing at night reduced to 14% (95% CI 8% to 20%). Similar activity budgets were achieved from the halter mounted sensors, but not those on a body harness. These results are consistent with previous studies directly observing daily activity of pasture-based sheep and can be applied in a variety of contexts to investigate animal health and welfare metrics e.g., to better understand the impact that young sheep can suffer when carrying even modest burdens of parasitic nematodes.

Animal tag technology keeps coming of age: an engineering perspective

Animal-borne tags (biologgers) have now become extremely sophisticated, recording data from multiple sensors at high frequencies for long periods and, as such, have become a powerful tool for behavioural ecologists and physiologists studying wild animals. But the design and implementation of these tags is not trivial because engineers have to maximize performance and ability to function under onerous conditions while minimizing tag mass and volume (footprint) to maximize the wellbeing of the animal carriers. We present some of the major issues faced by tag engineers and show how tag designers must accept compromises while maintaining systems that can answer the questions being posed. We also argue that basic understanding of engineering issues in tag design by biologists will help feedback to engineers to better tag construction but also reduce the likelihood that tag-deploying biologists will misunderstand their own results. Finally, we suggest that proper consideration of conventional technology together with new approaches will lead to further step changes in our understanding of wild-animal biology using smart tags. This article is part of the theme issue 'Measuring physiology in free-living animals (Part II)'.

Using Machine Learning for Remote Behaviour Classification-Verifying Acceleration Data to Infer Feeding Events in Free-Ranging Cheetahs

Behavioural studies of elusive wildlife species are challenging but important when they are threatened and involved in human-wildlife conflicts. Accelerometers (ACCs) and supervised machine learning algorithms (MLAs) are valuable tools to remotely determine behaviours. Here we used five captive cheetahs in Namibia to test the applicability of ACC data in identifying six behaviours by using six MLAs on data we ground-truthed by direct observations. We included two ensemble learning approaches and a probability threshold to improve prediction accuracy. We used the model to then identify the behaviours in four free-ranging cheetah males. Feeding behaviours identified by the model and matched with corresponding GPS clusters were verified with previously identified kill sites in the field. The MLAs and the two ensemble learning approaches in the captive cheetahs achieved precision (recall) ranging from 80.1% to 100.0% (87.3% to 99.2%) for resting, walking and trotting/running behaviour, from 74.4% to 81.6% (54.8% and 82.4%) for feeding behaviour and from 0.0% to 97.1% (0.0% and 56.2%) for drinking and grooming behaviour. The model application to the ACC data of the free-ranging cheetahs successfully identified all nine kill sites and 17 of the 18 feeding events of the two brother groups. We demonstrated that our behavioural model reliably detects feeding events of free-ranging cheetahs. This has useful applications for the determination of cheetah kill sites and helping to mitigate human-cheetah conflicts.

Quantifying finer-scale behaviours using self-organising maps (SOMs) to link accelerometery signatures with behavioural patterns in free-roaming terrestrial animals

Collecting quantitative information on animal behaviours is difficult, especially from cryptic species or species that alter natural behaviours under observation. Using harness-mounted tri-axial accelerometers free-roaming domestic cats (Felis Catus) we developed a methodology that can precisely classify finer-scale behaviours. We further tested the effect of a prey-protector device designed to reduce prey capture. We aligned accelerometer traces collected at 50 Hz with video files (60 fps) and labelled 12 individual behaviours, then trained a supervised machine-learning algorithm using Kohonen super self-organising maps (SOM). The SOM was able to predict individual behaviours with a ~ 99.6% overall accuracy, which was slightly better than for random forest estimates using the same dataset (98.9%). There was a significant effect of sample size, with precision and sensitivity decreasing rapidly below 2000 1-s observations. We were also able to detect a behaviour specific reduction in the predictability when cats were fitted with the prey-protector device indicating it altered biomechanical gait. Our results can be applied in movement ecology, zoology and conservation, where habitat specific movement performance between predators or prey may be critical to managing species of conservation significance, or in veterinary and agricultural fields, where early detection of movement pathologies can improve animal welfare.

Limitations of using surrogates for behaviour classification of accelerometer data: refining methods using random forest models in Caprids

BACKGROUND

Animal-attached devices can be used on cryptic species to measure their movement and behaviour, enabling unprecedented insights into fundamental aspects of animal ecology and behaviour. However, direct observations of subjects are often still necessary to translate biologging data accurately into meaningful behaviours. As many elusive species cannot easily be observed in the wild, captive or domestic surrogates are typically used to calibrate data from devices. However, the utility of this approach remains equivocal.

METHODS

Here, we assess the validity of using captive conspecifics, and phylogenetically-similar domesticated counterparts (surrogate species) for calibrating behaviour classification. Tri-axial accelerometers and tri-axial magnetometers were used with behavioural observations to build random forest models to predict the behaviours. We applied these methods using captive Alpine ibex (Capra ibex) and a domestic counterpart, pygmy goats (Capra aegagrus hircus), to predict the behaviour including terrain slope for locomotion behaviours of captive Alpine ibex.

RESULTS

Behavioural classification of captive Alpine ibex and domestic pygmy goats was highly accurate (> 98%). Model performance was reduced when using data split per individual, i.e., classifying behaviour of individuals not used to train models (mean ± sd = 56.1 ± 11%). Behavioural classifications using domestic counterparts, i.e., pygmy goat observations to predict ibex behaviour, however, were not sufficient to predict all behaviours of a phylogenetically similar species accurately (> 55%).

CONCLUSIONS

We demonstrate methods to refine the use of random forest models to classify behaviours of both captive and free-living animal species. We suggest there are two main reasons for reduced accuracy when using a domestic counterpart to predict the behaviour of a wild species in captivity; domestication leading to morphological differences and the terrain of the environment in which the animals were observed. We also identify limitations when behaviour is predicted in individuals that are not used to train models. Our results demonstrate that biologging device calibration needs to be conducted using: (i) with similar conspecifics, and (ii) in an area where they can perform behaviours on terrain that reflects that of species in the wild.

Behaviour Classification on Giraffes (Giraffa camelopardalis) Using Machine Learning Algorithms on Triaxial Acceleration Data of Two Commonly Used GPS Devices and Its Possible Application for Their Management and Conservation

Averting today's loss of biodiversity and ecosystem services can be achieved through conservation efforts, especially of keystone species. Giraffes (Giraffa camelopardalis) play an important role in sustaining Africa's ecosystems, but are 'vulnerable' according to the IUCN Red List since 2016. Monitoring an animal's behavior in the wild helps to develop and assess their conservation management. One mechanism for remote tracking of wildlife behavior is to attach accelerometers to animals to record their body movement. We tested two different commercially available high-resolution accelerometers, e-obs and Africa Wildlife Tracking (AWT), attached to the top of the heads of three captive giraffes and analyzed the accuracy of automatic behavior classifications, focused on the Random Forests algorithm. For both accelerometers, behaviors of lower variety in head and neck movements could be better predicted (i.e., feeding above eye level, mean prediction accuracy e-obs/AWT: 97.6%/99.7%; drinking: 96.7%/97.0%) than those with a higher variety of body postures (such as standing: 90.7-91.0%/75.2-76.7%; rumination: 89.6-91.6%/53.5-86.5%). Nonetheless both devices come with limitations and especially the AWT needs technological adaptations before applying it on animals in the wild. Nevertheless, looking at the prediction results, both are promising accelerometers for behavioral classification of giraffes. Therefore, these devices when applied to free-ranging animals, in combination with GPS tracking, can contribute greatly to the conservation of giraffes.

Hidden Markov models identify major movement modes in accelerometer and magnetometer data from four albatross species

BACKGROUND

Inertial measurement units (IMUs) with high-resolution sensors such as accelerometers are now used extensively to study fine-scale behavior in a wide range of marine and terrestrial animals. Robust and practical methods are required for the computationally-demanding analysis of the resulting large datasets, particularly for automating classification routines that construct behavioral time series and time-activity budgets. Magnetometers are used increasingly to study behavior, but it is not clear how these sensors contribute to the accuracy of behavioral classification methods. Development of effective  classification methodology is key to understanding energetic and life-history implications of foraging and other behaviors.

METHODS

We deployed accelerometers and magnetometers on four species of free-ranging albatrosses and evaluated the ability of unsupervised hidden Markov models (HMMs) to identify three major modalities in their behavior: 'flapping flight', 'soaring flight', and 'on-water'. The relative contribution of each sensor to classification accuracy was measured by comparing HMM-inferred states with expert classifications identified from stereotypic patterns observed in sensor data.

RESULTS

HMMs provided a flexible and easily interpretable means of classifying behavior from sensor data. Model accuracy was high overall (92%), but varied across behavioral states (87.6, 93.1 and 91.7% for 'flapping flight', 'soaring flight' and 'on-water', respectively). Models built on accelerometer data alone were as accurate as those that also included magnetometer data; however, the latter were useful for investigating slow and periodic behaviors such as dynamic soaring at a fine scale.

CONCLUSIONS

The use of IMUs in behavioral studies produces large data sets, necessitating the development of computationally-efficient methods to automate behavioral classification in order to synthesize and interpret underlying patterns. HMMs provide an accessible and robust framework for analyzing complex IMU datasets and comparing behavioral variation among taxa across habitats, time and space.

Assessment of Machine Learning Models to Identify Port Jackson Shark Behaviours Using Tri-Axial Accelerometers

Movement ecology has traditionally focused on the movements of animals over large time scales, but, with advancements in sensor technology, the focus can become increasingly fine scale. Accelerometers are commonly applied to quantify animal behaviours and can elucidate fine-scale (<2 s) behaviours. Machine learning methods are commonly applied to animal accelerometry data; however, they require the trial of multiple methods to find an ideal solution. We used tri-axial accelerometers (10 Hz) to quantify four behaviours in Port Jackson sharks (Heterodontus portusjacksoni): two fine-scale behaviours (<2 s)-(1) vertical swimming and (2) chewing as proxy for foraging, and two broad-scale behaviours (>2 s-mins)-(3) resting and (4) swimming. We used validated data to calculate 66 summary statistics from tri-axial accelerometry and assessed the most important features that allowed for differentiation between the behaviours. One and two second epoch testing sets were created consisting of 10 and 20 samples from each behaviour event, respectively. We developed eight machine learning models to assess their overall accuracy and behaviour-specific accuracy (one classification tree, five ensemble learners and two neural networks). The support vector machine model classified the four behaviours better when using the longer 2 s time epoch (F-measure 89%; macro-averaged F-measure: 90%). Here, we show that this support vector machine (SVM) model can reliably classify both fine- and broad-scale behaviours in Port Jackson sharks.

Challenges of machine learning model validation using correlated behaviour data: Evaluation of cross-validation strategies and accuracy measures

Automated monitoring of the movements and behaviour of animals is a valuable research tool. Recently, machine learning tools were applied to many species to classify units of behaviour. For the monitoring of wild species, collecting enough data for training models might be problematic, thus we examine how machine learning models trained on one species can be applied to another closely related species with similar behavioural conformation. We contrast two ways to calculate accuracies, termed here as overall and threshold accuracy, because the field has yet to define solid standards for reporting and measuring classification performances. We measure 21 dogs and 7 wolves, and find that overall accuracies are between 51 and 60% for classifying 8 behaviours (lay, sit, stand, walk, trot, run, eat, drink) when training and testing data are from the same species and between 41 and 51% when training and testing is cross-species. We show that using data from dogs to predict the behaviour of wolves is feasible. We also show that optimising the model for overall accuracy leads to similar overall and threshold accuracies, while optimizing for threshold accuracy leads to threshold accuracies well above 80%, but yielding very low overall accuracies, often below the chance level. Moreover, we show that the most common method for dividing the data between training and testing data (random selection of test data) overestimates the accuracy of models when applied to data of new specimens. Consequently, we argue that for the most common goals of animal behaviour recognition overall accuracy should be the preferred metric. Considering, that often the goal is to collect movement data without other methods of observation, we argue that training data and testing data should be divided by individual and not randomly.

Behavioural inference from signal processing using animal-borne multi-sensor loggers: a novel solution to extend the knowledge of sea turtle ecology

The identification of sea turtle behaviours is a prerequisite to predicting the activities and time-budget of these animals in their natural habitat over the long term. However, this is hampered by a lack of reliable methods that enable the detection and monitoring of certain key behaviours such as feeding. This study proposes a combined approach that automatically identifies the different behaviours of free-ranging sea turtles through the use of animal-borne multi-sensor recorders (accelerometer, gyroscope and time-depth recorder), validated by animal-borne video-recorder data. We show here that the combination of supervised learning algorithms and multi-signal analysis tools can provide accurate inferences of the behaviours expressed, including feeding and scratching behaviours that are of crucial ecological interest for sea turtles. Our procedure uses multi-sensor miniaturized loggers that can be deployed on free-ranging animals with minimal disturbance. It provides an easily adaptable and replicable approach for the long-term automatic identification of the different activities and determination of time-budgets in sea turtles. This approach should also be applicable to a broad range of other species and could significantly contribute to the conservation of endangered species by providing detailed knowledge of key animal activities such as feeding, travelling and resting.

Identifying Sheep Activity from Tri-Axial Acceleration Signals Using a Moving Window Classification Model

Behaviour is a useful indicator of an individual animal’s overall wellbeing. There is widespread agreement that measuring and monitoring individual behaviour autonomously can provide valuable opportunities to trigger and refine on-farm management decisions. Conventionally, this has required visual observation of animals across a set time period. Technological advancements, such as animal-borne accelerometers, are offering 24/7 monitoring capability. Accelerometers have been used in research to quantify animal behaviours for a number of years. Now, technology and software developers, and more recently decision support platform providers, are integrating to offer commercial solutions for the extensive livestock industries. For these systems to function commercially, data must be captured, processed and analysed in sync with data acquisition. Practically, this requires a continuous stream of data or a duty cycled data segment and, from an analytics perspective, the application of moving window algorithms to derive the required classification. The aim of this study was to evaluate the application of a ‘clean state’ moving window behaviour state classification algorithm applied to 3, 5 and 10 second duration segments of data (including behaviour transitions), to categorise data emanating from collar, leg and ear mounted accelerometers on five Merino ewes. The model was successful at categorising grazing, standing, walking and lying behaviour classes with varying sensitivity, and no significant difference in model accuracy was observed between the three moving window lengths. The accuracy in identifying behaviour classes was highest for the ear-mounted sensor (86%–95%), followed by the collar-mounted sensor (67%–88%) and leg-mounted sensor (48%–94%). Between-sheep variations in classification accuracy confirm the sensor orientation is an important source of variation in all deployment modes. This research suggests a moving window classifier is capable of segregating continuous accelerometer signals into exclusive behaviour classes and may provide an appropriate data processing framework for commercial deployments.

Monitoring canid scent marking in space and time using a biologging and machine learning approach

For canid species, scent marking plays a critical role in territoriality, social dynamics, and reproduction. However, due in part to human dependence on vision as our primary sensory modality, research on olfactory communication is hampered by a lack of tractable methods. In this study, we leverage a powerful biologging approach, using accelerometers in concert with GPS loggers to monitor and describe scent-marking events in time and space. We performed a validation experiment with domestic dogs, monitoring them by video concurrently with the novel biologging approach. We attached an accelerometer to the pelvis of 31 dogs (19 males and 12 females), detecting raised-leg and squat posture urinations by monitoring the change in device orientation. We then deployed this technique to describe the scent marking activity of 3 guardian dogs as they defend livestock from coyote depredation in California, providing an example use-case for the technique. During validation, the algorithm correctly classified 92% of accelerometer readings. High performance was partly due to the conspicuous signatures of archetypal raised-leg postures in the accelerometer data. Accuracy did not vary with the weight, age, and sex of the dogs, resulting in a method that is broadly applicable across canid species’ morphologies. We also used models trained on each individual to detect scent marking of others to emulate the use of captive surrogates for model training. We observed no relationship between the similarity in body weight between the dog pairs and the overall accuracy of predictions, although models performed best when trained and tested on the same individual. We discuss how existing methods in the field of movement ecology can be extended to use this exciting new data type. This paper represents an important first step in opening new avenues of research by leveraging the power of modern-technologies and machine-learning to this field.

Behavioral classification of low-frequency acceleration and temperature data from a free-ranging small mammal

The miniaturization and affordability of new technology is driving a biologging revolution in wildlife ecology with use of animal-borne data logging devices. Among many new biologging technologies, accelerometers are emerging as key tools for continuously recording animal behavior. Yet a critical, but under-acknowledged consideration in biologging is the trade-off between sampling rate and sampling duration, created by battery- (or memory-) related sampling constraints. This is especially acute among small animals, causing most researchers to sample at high rates for very limited durations. Here, we show that high accuracy in behavioral classification is achievable when pairing low-frequency acceleration recordings with temperature. We conducted 84 hr of direct behavioral observations on 67 free-ranging red squirrels (200-300 g) that were fitted with accelerometers (2 g) recording tri-axial acceleration and temperature at 1 Hz. We then used a random forest algorithm and a manually created decision tree, with variable sampling window lengths, to associate observed behavior with logger recorded acceleration and temperature. Finally, we assessed the accuracy of these different classifications using an additional 60 hr of behavioral observations, not used in the initial classification. The accuracy of the manually created decision tree classification using observational data varied from 70.6% to 91.6% depending on the complexity of the tree, with increasing accuracy as complexity decreased. Short duration behavior like running had lower accuracy than long-duration behavior like feeding. The random forest algorithm offered similarly high overall accuracy, but the manual decision tree afforded the flexibility to create a hierarchical tree, and to adjust sampling window length for behavioral states with varying durations. Low frequency biologging of acceleration and temperature allows accurate behavioral classification of small animals over multi-month sampling durations. Nevertheless, low sampling rates impose several important limitations, especially related to assessing the classification accuracy of short duration behavior.

High accuracy at low frequency: detailed behavioural classification from accelerometer data

Accelerometers are a valuable tool for studying animal behaviour and physiology where direct observation is unfeasible. However, giving biological meaning to multivariate acceleration data is challenging. Here, we describe a method that reliably classifies a large number of behaviours using tri-axial accelerometer data collected at the low sampling frequency of 1 Hz, using the dingo (Canis dingo) as an example. We used out-of-sample validation to compare the predictive performance of four commonly used classification models (random forest, k-nearest neighbour, support vector machine, and naïve Bayes). We tested the importance of predictor variable selection and moving window size for the classification of each behaviour and overall model performance. Random forests produced the highest out-of-sample classification accuracy, with our best-performing model predicting 14 behaviours with a mean accuracy of 87%. We also investigated the relationship between overall dynamic body acceleration (ODBA) and the activity level of each behaviour, given the increasing use of ODBA in ecophysiology as a proxy for energy expenditure. ODBA values for our four 'high activity' behaviours were significantly greater than all other behaviours, with an overall positive trend between ODBA and intensity of movement. We show that a random forest model of relatively low complexity can mitigate some major challenges associated with establishing meaningful ecological conclusions from acceleration data. Our approach has broad applicability to free-ranging terrestrial quadrupeds of comparable size. Our use of a low sampling frequency shows potential for deploying accelerometers over extended time periods, enabling the capture of invaluable behavioural and physiological data across different ontogenies.

Predicting Lameness in Sheep Activity Using Tri-Axial Acceleration Signals

Simple Summary

Monitoring livestock farmed under extensive conditions is challenging and this is particularly difficult when observing animal behaviour at an individual level. Lameness is a disease symptom that has traditionally relied on visual inspection to detect those animals with an abnormal walking pattern. More recently, accelerometer sensors have been used in other livestock industries to detect lame animals. These devices are able to record changes in activity intensity, allowing us to differentiate between a grazing, walking, and resting animal. Using these on-animal sensors, grazing, standing, walking, and lame walking were accurately detected from an ear attached sensor. With further development, this classification algorithm could be linked with an automatic livestock monitoring system to provide real time information on individual health status, something that is practically not possible under current extensive livestock production systems.

Abstract

Lameness is a clinical symptom associated with a number of sheep diseases around the world, having adverse effects on weight gain, fertility, and lamb birth weight, and increasing the risk of secondary diseases. Current methods to identify lame animals rely on labour intensive visual inspection. The aim of this current study was to determine the ability of a collar, leg, and ear attached tri-axial accelerometer to discriminate between sound and lame gait movement in sheep. Data were separated into 10 s mutually exclusive behaviour epochs and subjected to Quadratic Discriminant Analysis (QDA). Initial analysis showed the high misclassification of lame grazing events with sound grazing and standing from all deployment modes. The final classification model, which included lame walking and all sound activity classes, yielded a prediction accuracy for lame locomotion of 82%, 35%, and 87% for the ear, collar, and leg deployments, respectively. Misclassification of sound walking with lame walking within the leg accelerometer dataset highlights the superiority of an ear mode of attachment for the classification of lame gait characteristics based on time series accelerometer data.

Development and application of a machine learning algorithm for classification of elasmobranch behaviour from accelerometry data

Discerning behaviours of free-ranging animals allows for quantification of their activity budget, providing important insight into ecology. Over recent years, accelerometers have been used to unveil the cryptic lives of animals. The increased ability of accelerometers to store large quantities of high resolution data has prompted a need for automated behavioural classification. We assessed the performance of several machine learning (ML) classifiers to discern five behaviours performed by accelerometer-equipped juvenile lemon sharks (Negaprion brevirostris) at Bimini, Bahamas (25°44'N, 79°16'W). The sharks were observed to exhibit chafing, burst swimming, headshaking, resting and swimming in a semi-captive environment and these observations were used to ground-truth data for ML training and testing. ML methods included logistic regression, an artificial neural network, two random forest models, a gradient boosting model and a voting ensemble (VE) model, which combined the predictions of all other (base) models to improve classifier performance. The macro-averaged F-measure, an indicator of classifier performance, showed that the VE model improved overall classification (F-measure 0.88) above the strongest base learner model, gradient boosting (0.86). To test whether the VE model provided biologically meaningful results when applied to accelerometer data obtained from wild sharks, we investigated headshaking behaviour, as a proxy for prey capture, in relation to the variables: time of day, tidal phase and season. All variables were significant in predicting prey capture, with predations most likely to occur during early evening and less frequently during the dry season and high tides. These findings support previous hypotheses from sporadic visual observations.

Seeing It All: Evaluating Supervised Machine Learning Methods for the Classification of Diverse Otariid Behaviours

Constructing activity budgets for marine animals when they are at sea and cannot be directly observed is challenging, but recent advances in bio-logging technology offer solutions to this problem. Accelerometers can potentially identify a wide range of behaviours for animals based on unique patterns of acceleration. However, when analysing data derived from accelerometers, there are many statistical techniques available which when applied to different data sets produce different classification accuracies. We investigated a selection of supervised machine learning methods for interpreting behavioural data from captive otariids (fur seals and sea lions). We conducted controlled experiments with 12 seals, where their behaviours were filmed while they were wearing 3-axis accelerometers. From video we identified 26 behaviours that could be grouped into one of four categories (foraging, resting, travelling and grooming) representing key behaviour states for wild seals. We used data from 10 seals to train four predictive classification models: stochastic gradient boosting (GBM), random forests, support vector machine using four different kernels and a baseline model: penalised logistic regression. We then took the best parameters from each model and cross-validated the results on the two seals unseen so far. We also investigated the influence of feature statistics (describing some characteristic of the seal), testing the models both with and without these. Cross-validation accuracies were lower than training accuracy, but the SVM with a polynomial kernel was still able to classify seal behaviour with high accuracy (>70%). Adding feature statistics improved accuracies across all models tested. Most categories of behaviour -resting, grooming and feeding—were all predicted with reasonable accuracy (52–81%) by the SVM while travelling was poorly categorised (31–41%). These results show that model selection is important when classifying behaviour and that by using animal characteristics we can strengthen the overall accuracy.

An exploratory clustering approach for extracting stride parameters from tracking collars on free-ranging wild animals

Changes in stride frequency and length with speed are key parameters in animal locomotion research. They are commonly measured in a laboratory on a treadmill or by filming trained captive animals. Here, we show that a clustering approach can be used to extract these variables from data collected by a tracking collar containing a GPS module and tri-axis accelerometers and gyroscopes. The method enables stride parameters to be measured during free-ranging locomotion in natural habitats. As it does not require labelled data, it is particularly suitable for use with difficult to observe animals. The method was tested on large data sets collected from collars on free-ranging lions and African wild dogs and validated using a domestic dog.

The private life of echidnas: using accelerometry and GPS to examine field biomechanics and assess the ecological impact of a widespread, semi-fossorial monotreme

The short-beaked echidna (Tachyglossus aculeatus) is a monotreme and therefore provides a unique combination of phylogenetic history, morphological differentiation and ecological specialisation for a mammal. The echidna has a unique appendicular skeleton, a highly specialised myrmecophagous lifestyle and a mode of locomotion that is neither typically mammalian nor reptilian, but has aspects of both lineages. We therefore were interested in the interactions of locomotor biomechanics, ecology and movements for wild, free-living short-beaked echidnas. To assess locomotion in its complex natural environment, we attached both GPS and accelerometer loggers to the back of echidnas in both spring and summer. We found that the locomotor biomechanics of echidnas is unique, with lower stride length and stride frequency than reported for similar-sized mammals. Speed modulation is primarily accomplished through changes in stride frequency, with a mean of 1.39 Hz and a maximum of 2.31 Hz. Daily activity period was linked to ambient air temperature, which restricted daytime activity during the hotter summer months. Echidnas had longer activity periods and longer digging bouts in spring compared with summer. In summer, echidnas had higher walking speeds than in spring, perhaps because of the shorter time suitable for activity. Echidnas spent, on average, 12% of their time digging, which indicates their potential to excavate up to 204 m3 of soil a year. This information highlights the important contribution towards ecosystem health, via bioturbation, of this widespread Australian monotreme.

Using accelerometers to remotely and automatically characterize behavior in small animals

Activity budgets in wild animals are challenging to measure via direct observation because data collection is time consuming and observer effects are potentially confounding. Although tri-axial accelerometers are increasingly employed for this purpose, their application in small-bodied animals has been limited by weight restrictions. Additionally, accelerometers engender novel complications, as a system is needed to reliably map acceleration to behaviors. In this study, we describe newly developed, tiny acceleration-logging devices (1.5-2.5 g) and use them to characterize behavior in two chipmunk species. We collected paired accelerometer readings and behavioral observations from captive individuals. We then employed techniques from machine learning to develop an automatic system for coding accelerometer readings into behavioral categories. Finally, we deployed and recovered accelerometers from free-living, wild chipmunks. This is the first time to our knowledge that accelerometers have been used to generate behavioral data for small-bodied (<100 g), free-living mammals.

Interpreting behaviors from accelerometry: a method combining simplicity and objectivity

Quantifying the behavior of motile, free-ranging animals is difficult. The accelerometry technique offers a method for recording behaviors but interpretation of the data is not straightforward. To date, analysis of such data has either involved subjective, study-specific assignments of behavior to acceleration data or the use of complex analyses based on machine learning. Here, we present a method for automatically classifying acceleration data to represent discrete, coarse-scale behaviors. The method centers on examining the shape of histograms of basic metrics readily derived from acceleration data to objectively determine threshold values by which to separate behaviors. Through application of this method to data collected on two distinct species with greatly differing behavioral repertoires, kittiwakes, and humans, the accuracy of this approach is demonstrated to be very high, comparable to that reported for other automated approaches already published. The method presented offers an alternative to existing methods as it uses biologically grounded arguments to distinguish behaviors, it is objective in determining values by which to separate these behaviors, and it is simple to implement, thus making it potentially widely applicable. The R script coding the method is provided.

Step by step: reconstruction of terrestrial animal movement paths by dead-reckoning

BACKGROUND

Research on wild animal ecology is increasingly employing GPS telemetry in order to determine animal movement. However, GPS systems record position intermittently, providing no information on latent position or track tortuosity. High frequency GPS have high power requirements, which necessitates large batteries (often effectively precluding their use on small animals) or reduced deployment duration. Dead-reckoning is an alternative approach which has the potential to 'fill in the gaps' between less resolute forms of telemetry without incurring the power costs. However, although this method has been used in aquatic environments, no explicit demonstration of terrestrial dead-reckoning has been presented.

RESULTS

We perform a simple validation experiment to assess the rate of error accumulation in terrestrial dead-reckoning. In addition, examples of successful implementation of dead-reckoning are given using data from the domestic dog Canus lupus, horse Equus ferus, cow Bos taurus and wild badger Meles meles.

CONCLUSIONS

This study documents how terrestrial dead-reckoning can be undertaken, describing derivation of heading from tri-axial accelerometer and tri-axial magnetometer data, correction for hard and soft iron distortions on the magnetometer output, and presenting a novel correction procedure to marry dead-reckoned paths to ground-truthed positions. This study is the first explicit demonstration of terrestrial dead-reckoning, which provides a workable method of deriving the paths of animals on a step-by-step scale. The wider implications of this method for the understanding of animal movement ecology are discussed.

The Use of Acceleration to Code for Animal Behaviours; A Case Study in Free-Ranging Eurasian Beavers Castor fiber

Recent technological innovations have led to the development of miniature, accelerometer-containing electronic loggers which can be attached to free-living animals. Accelerometers provide information on both body posture and dynamism which can be used as descriptors to define behaviour. We deployed tri-axial accelerometer loggers on 12 free-ranging Eurasian beavers Castor fiber in the county of Telemark, Norway, and on four captive beavers (two Eurasian beavers and two North American beavers C. canadensis) to corroborate acceleration signals with observed behaviours. By using random forests for classifying behavioural patterns of beavers from accelerometry data, we were able to distinguish seven behaviours; standing, walking, swimming, feeding, grooming, diving and sleeping. We show how to apply the use of acceleration to determine behaviour, and emphasise the ease with which this non-invasive method can be implemented. Furthermore, we discuss the strengths and weaknesses of this, and the implementation of accelerometry on animals, illustrating limitations, suggestions and solutions. Ultimately, this approach may also serve as a template facilitating studies on other animals with similar locomotor modes and deliver new insights into hitherto unknown aspects of behavioural ecology.

Love thy neighbour: automatic animal behavioural classification of acceleration data using the K-nearest neighbour algorithm

Researchers hoping to elucidate the behaviour of species that aren't readily observed are able to do so using biotelemetry methods. Accelerometers in particular are proving particularly effective and have been used on terrestrial, aquatic and volant species with success. In the past, behavioural modes were detected in accelerometer data through manual inspection, but with developments in technology, modern accelerometers now record at frequencies that make this impractical. In light of this, some researchers have suggested the use of various machine learning approaches as a means to classify accelerometer data automatically. We feel uptake of this approach by the scientific community is inhibited for two reasons; 1) Most machine learning algorithms require selection of summary statistics which obscure the decision mechanisms by which classifications are arrived, and 2) they are difficult to implement without appreciable computational skill. We present a method which allows researchers to classify accelerometer data into behavioural classes automatically using a primitive machine learning algorithm, k-nearest neighbour (KNN). Raw acceleration data may be used in KNN without selection of summary statistics, and it is easily implemented using the freeware program R. The method is evaluated by detecting 5 behavioural modes in 8 species, with examples of quadrupedal, bipedal and volant species. Accuracy and Precision were found to be comparable with other, more complex methods. In order to assist in the application of this method, the script required to run KNN analysis in R is provided. We envisage that the KNN method may be coupled with methods for investigating animal position, such as GPS telemetry or dead-reckoning, in order to implement an integrated approach to movement ecology research.

Optimizing acceleration-based ethograms: the use of variable-time versus fixed-time segmentation

BACKGROUND

Animal-borne accelerometers measure body orientation and movement and can thus be used to classify animal behaviour. To univocally and automatically analyse the large volume of data generated, we need classification models. An important step in the process of classification is the segmentation of acceleration data, i.e. the assignment of the boundaries between different behavioural classes in a time series. So far, analysts have worked with fixed-time segments, but this may weaken the strength of the derived classification models because transitions of behaviour do not necessarily coincide with boundaries of the segments. Here we develop random forest automated supervised classification models either built on variable-time segments generated with a so-called 'change-point model', or on fixed-time segments, and compare for eight behavioural classes the classification performance. The approach makes use of acceleration data measured in eight free-ranging crab plovers Dromas ardeola.

RESULTS

Useful classification was achieved by both the variable-time and fixed-time approach for flying (89% vs. 91%, respectively), walking (88% vs. 87%) and body care (68% vs. 72%). By using the variable-time segment approach, significant gains in classification performance were obtained for inactive behaviours (95% vs. 92%) and for two major foraging activities, i.e. handling (84% vs. 77%) and searching (78% vs. 67%). Attacking a prey and pecking were never accurately classified by either method.

CONCLUSION

Acceleration-based behavioural classification can be optimized using a variable-time segmentation approach. After implementing variable-time segments to our sample data, we achieved useful levels of classification performance for almost all behavioural classes. This enables behaviour, including motion, to be set in known spatial contexts, and the measurement of behavioural time-budgets of free-living birds with unprecedented coverage and precision. The methods developed here can be easily adopted in other studies, but we emphasize that for each species and set of questions, the presented string of work steps should be run through.

A combination of gyroscope and accelerometer for identifying alternative feeding behaviours in fish

We examined whether we could identify the feeding behaviours of the trophic generalist fish Epinephelus ongus on different prey types (crabs and fish) using a data-logger that incorporated a 3-axis gyroscope and a 3-axis accelerometer. Feeding behaviours and other burst behaviours, including escape responses, intraspecific interactions, and routine movements, were recorded from six E. ongus individuals using data-loggers sampling at 200 Hz, and were validated by simultaneously recorded video images. For each data-logger record, we extracted 5 seconds of data when any of the 3-axis accelerations exceeded absolute 2.0 G, to capture all feeding behaviours and other burst behaviours. Each feeding behaviour was then identified using a combination of parameters that were derived from the extracted data. Using decision trees with the parameters, high true identification rates (87.5% for both feeding behaviours) with low false identification rates (5% for crab-eating and 6.3% for fish-eating) were achieved for both feeding behaviours.