Movement Activity Based Classification of Animal Behaviour with an Application to Data from Cheetah (Acinonyx jubatus)

We propose a new method, based on machine learning techniques, for the analysis of a combination of continuous data from dataloggers and a sampling of contemporaneous behaviour observations. This data combination provides an opportunity for biologists to study behaviour at a previously unknown level of detail and accuracy; however, continuously recorded data are of little use unless the resulting large volumes of raw data can be reliably translated into actual behaviour. We address this problem by applying a Support Vector Machine and a Hidden-Markov Model that allows us to classify an animal''s behaviour using a small set of field observations to calibrate continuously recorded activity data. Such classified data can be applied quantitatively to the behaviour of animals over extended periods and at times during which observation is difficult or impossible. We demonstrate the usefulness of the method by applying it to data from six cheetah (Acinonyx jubatus) in the Okavango Delta, Botswana. Cumulative activity data scores were recorded every five minutes by accelerometers embedded in GPS radio-collars for around one year on average. Direct behaviour sampling of each of the six cheetah were collected in the field for comparatively short periods. Using this approach we are able to classify each five minute activity score into a set of three key behaviour (feeding, mobile and stationary), creating a continuous behavioural sequence for the entire period for which the collars were deployed. Evaluation of our classifier with cross-validation shows the accuracy to be , but that the accuracy for individual classes is reduced with decreasing sample size of direct observations. We demonstrate how these processed data can be used to study behaviour identifying seasonal and gender differences in daily activity and feeding times. Results given here are unlike any that could be obtained using traditional approaches in both accuracy and detail.     

Precocial development of locomotor performance in a ground-dwelling bird (Alectoris chukar): negotiating a three-dimensional terrestrial environment

Developing animals are particularly vulnerable to predation. Hence, precocial young of many taxa develop predator escape performance that rivals that of adults. Ontogenetically unique among vertebrates, birds transition from hind limb to forelimb dependence for escape behaviours, so developmental investment for immediate gains in running performance may impair flight performance later. Here, in a three-dimensional kinematic study of developing birds performing pre-flight flapping locomotor behaviours, wing-assisted incline running (WAIR) and a newly described behaviour, controlled flapping descent (CFD), we define three stages of locomotor ontogeny in a model gallinaceous bird (Alectoris chukar). In stage I (1–7 days post-hatching (dph)) birds crawl quadrupedally during ascents, and their flapping fails to reduce their acceleration during aerial descents. Stage II (8–19 dph) birds use symmetric wing beats during WAIR, and in CFD significantly reduce acceleration while controlling body pitch to land on their feet. In stage III (20 dph to adults), birds are capable of vertical WAIR and level-powered flight. In contrast to altricial species, which first fly when nearly at adult mass, we show that in a precocial bird the major requirements for flight (i.e. high power output, wing control and wing size) convene by around 8 dph (at ca 5% of adult mass) and yield significant gains in escape performance: immature chukars can fly by 20 dph, at only about 12 per cent of adult mass.     

Classification of broiler behaviours using triaxial accelerometer and machine learning

Understanding broiler behaviours provides important implications for animal well-being and farm management. The objectives of this study were to classify specific broiler behaviours by analysing data from wearable accelerometers using two machine learning models, K-Nearest Neighbour (KNN) and Support Vector Machine (SVM). Lightweight triaxial accelerometers were used to record accelerations of nine 7-week-old broilers at a sampling frequency of 40 Hz. A total of 261.6-min data were labelled for four behaviours – walking, resting, feeding and drinking. Instantaneous motion features including magnitude area, vector magnitude, movement variation, energy, and entropy were extracted and stored in a dataset which was then segmented by one of the six window lengths (1, 3, 5, 7, 10 and 20 s) with 50% overlap between consecutive windows. The mean, variation, SD, minimum and maximum of each instantaneous motion feature and two-way correlations of acceleration data were calculated within each window, yielding a total of 43 statistic features for training and testing of machine learning models. Performance of the models was evaluated using pure behaviour datasets (single behaviour type per dataset) and continuous behaviour datasets (continuous recording that involved multiple behaviour types per dataset). For pure behaviour datasets, both KNN and SVM models showed high sensitivities in classifying broiler resting (87% and 85%, respectively) and walking (99% and 99%, respectively). The accuracies of SVM were higher than KNN in differentiating feeding (88% and 75%, respectively) and drinking (83% and 62%, respectively) behaviours. Sliding window with 1-s length yielded the best performance for classifying continuous behaviour datasets. The performance of classification model generally improved as more birds were included for training. In conclusion, classification of specific broiler behaviours can be achieved by recording bird triaxial accelerations and analysing acceleration data through machine learning. Performances of different machine learning models differ in classifying specific broiler behaviours.     

Movement Behaviour of Traditionally Managed Cattle in the Eastern Province of Zambia Captured Using Two-Dimensional Motion Sensors

Two-dimensional motion sensors use electronic accelerometers to record the lying, standing and walking activity of cattle. Movement behaviour data collected automatically using these sensors over prolonged periods of time could be of use to stakeholders making management and disease control decisions in rural sub-Saharan Africa leading to potential improvements in animal health and production. Motion sensors were used in this study with the aim of monitoring and quantifying the movement behaviour of traditionally managed Angoni cattle in Petauke District in the Eastern Province of Zambia. This study was designed to assess whether motion sensors were suitable for use on traditionally managed cattle in two veterinary camps in Petauke District in the Eastern Province of Zambia. In each veterinary camp, twenty cattle were selected for study. Each animal had a motion sensor placed on its hind leg to continuously measure and record its movement behaviour over a two week period. Analysing the sensor data using principal components analysis (PCA) revealed that the majority of variability in behaviour among studied cattle could be attributed to their behaviour at night and in the morning. The behaviour at night was markedly different between veterinary camps; while differences in the morning appeared to reflect varying behaviour across all animals. The study results validate the use of such motion sensors in the chosen setting and highlight the importance of appropriate data summarisation techniques to adequately describe and compare animal movement behaviours if association to other factors, such as location, breed or health status are to be assessed.     

Classifying movement behaviour in relation to environmental conditions using hidden Markov models

1.  Linking the movement and behaviour of animals to their environment is a central problem in ecology. Through the use of electronic tagging and tracking (ETT), collection of in situ data from free-roaming animals is now commonplace, yet statistical approaches enabling direct relation of movement observations to environmental conditions are still in development.
2.  In this study, we examine the hidden Markov model (HMM) for behavioural analysis of tracking data. HMMs allow for prediction of latent behavioural states while directly accounting for the serial dependence prevalent in ETT data. Updating the probability of behavioural switches with tag or remote-sensing data provides a statistical method that links environmental data to behaviour in a direct and integrated manner.
3.  It is important to assess the reliability of state categorization over the range of time-series lengths typically collected from field instruments and when movement behaviours are similar between movement states. Simulation with varying lengths of times series data and contrast between average movements within each state was used to test the HMMs ability to estimate movement parameters.
4.  To demonstrate the methods in a realistic setting, the HMMs were used to categorize resident and migratory phases and the relationship between movement behaviour and ocean temperature using electronic tagging data from southern bluefin tuna ( Thunnus maccoyii ). Diagnostic tools to evaluate the suitability of different models and inferential methods for investigating differences in behaviour between individuals are also demonstrated.     

State-space models of individual animal movement

Detailed observation of the movement of individual animals offers the potential to understand spatial population processes as the ultimate consequence of individual behaviour, physiological constraints and fine-scale environmental influences. However, movement data from individuals are intrinsically stochastic and often subject to severe observation error. Linking such complex data to dynamical models of movement is a major challenge for animal ecology. Here, we review a statistical approach, state-space modelling, which involves changing how we analyse movement data and draw inferences about the behaviours that shape it. The statistical robustness and predictive ability of state-space models make them the most promising avenue towards a new type of movement ecology that fuses insights from the study of animal behaviour, biogeography and spatial population dynamics.     

Extending the Functionality of Behavioural Change-Point Analysis with k-Means Clustering: A Case Study with the Little Penguin (Eudyptula minor)

We present a simple framework for classifying mutually exclusive behavioural states within the geospatial lifelines of animals. This method involves use of three sequentially applied statistical procedures: (1) behavioural change point analysis to partition movement trajectories into discrete bouts of same-state behaviours, based on abrupt changes in the spatio-temporal autocorrelation structure of movement parameters; (2) hierarchical multivariate cluster analysis to determine the number of different behavioural states; and (3) k-means clustering to classify inferred bouts of same-state location observations into behavioural modes. We demonstrate application of the method by analysing synthetic trajectories of known ‘artificial behaviours’ comprised of different correlated random walks, as well as real foraging trajectories of little penguins (Eudyptula minor) obtained by global-positioning-system telemetry. Our results show that the modelling procedure correctly classified 92.5% of all individual location observations in the synthetic trajectories, demonstrating reasonable ability to successfully discriminate behavioural modes. Most individual little penguins were found to exhibit three unique behavioural states (resting, commuting/active searching, area-restricted foraging), with variation in the timing and locations of observations apparently related to ambient light, bathymetry, and proximity to coastlines and river mouths. Addition of k-means clustering extends the utility of behavioural change point analysis, by providing a simple means through which the behaviours inferred for the location observations comprising individual movement trajectories can be objectively classified.     

Measuring Energetics and Behaviour Using Accelerometry in Cane Toads Bufo marinus

Cane toads Bufo marinus were introduced to Australia as a control agent but now have a rapidly progressing invasion front and damage new habitats they enter. Predictive models that can give expansion rates as functions of energy supply and feeding ground distribution could help to maximise control efficiency but to date no study has measured rates of field energy expenditure in an amphibian. In the present study we used the accelerometry technique to generate behavioural time budgets and, through the derivation of ODBA (overall dynamic body acceleration), to obtain estimates of energetics in free ranging cane toads. This represents the first time that accelerometers have been used to not only quantify the behaviour of animals but also assign to those behaviours rates of energy expenditure. Firstly, laboratory calibrations between ODBA and metabolic rate were obtained and used to generate a common prediction equation for the subject toads (R² = 0.74). Furthermore, acceleration data recorded during different behaviours was studied to ascertain threshold values for objectively defining behaviour categories. Importantly, while subsequent accelerometer field deployments were relatively short they agreed with previous studies on the proportion of time that cane toads locomote yet suggest that the metabolic rate of cane toads in the wild may sometimes be considerably higher than might be assumed based on data for other species.     

Identifying spawning events of greater amberjack using accelerometers

Direct observations of spawning events of large free-swimming migratory fish are difficult in nature. However, behavioural changes specific to spawning events of fish species, i.e. swimming depth or acceleration of body motion using data loggers attached to the body, have successfully been used to identify spawning. In this study, we observed the spawning behaviour in an experimental tank of a pair of greater amberjack Seriola dumerili, which were matured by injection of gonadotropin, by attaching an acceleration data logger, an accelerometer. The male and female were observed to vibrate their bodies during the moment of spawning. Analysis of the acceleration data showed the simultaneous appearance of a dominant cycle (cycle indicating the highest amplitude in 1?s) under 0.4?s continuing over 7?sand dominant amplitude over 0.4 indicating the time of the spawning in both sexes. These characteristics are useful for the development of an algorithm to detect the spawning behaviour, allowing us to develop a new data logger with an on-board algorithm for detection of spawning behaviour. Before development of this system, we need to develop new algorithms for separation of spawning behaviour from other similar large motion behaviours, particularly from feeding and prey avoidance behaviour.     

Deriving Animal Behaviour from High-Frequency GPS: Tracking Cows in Open and Forested Habitat

The increasing spatiotemporal accuracy of Global Navigation Satellite Systems (GNSS) tracking systems opens the possibility to infer animal behaviour from tracking data. We studied the relationship between high-frequency GNSS data and behaviour, aimed at developing an easily interpretable classification method to infer behaviour from location data. Behavioural observations were carried out during tracking of cows (Bos Taurus) fitted with high-frequency GPS (Global Positioning System) receivers. Data were obtained in an open field and forested area, and movement metrics were calculated for 1 min, 12 s and 2 s intervals. We observed four behaviour types (Foraging, Lying, Standing and Walking). We subsequently used Classification and Regression Trees to classify the simultaneously obtained GPS data as these behaviour types, based on distances and turning angles between fixes. GPS data with a 1 min interval from the open field was classified correctly for more than 70% of the samples. Data from the 12 s and 2 s interval could not be classified successfully, emphasizing that the interval should be long enough for the behaviour to be defined by its characteristic movement metrics. Data obtained in the forested area were classified with a lower accuracy (57%) than the data from the open field, due to a larger positional error of GPS locations and differences in behavioural performance influenced by the habitat type. This demonstrates the importance of understanding the relationship between behaviour and movement metrics, derived from GNSS fixes at different frequencies and in different habitats, in order to successfully infer behaviour. When spatially accurate location data can be obtained, behaviour can be inferred from high-frequency GNSS fixes by calculating simple movement metrics and using easily interpretable decision trees. This allows for the combined study of animal behaviour and habitat use based on location data, and might make it possible to detect deviations in behaviour at the individual level.     

A Generic Individual-Based Spatially Explicit Model as a Novel Tool for Investigating Insect-Plant Interactions: A Case Study of the Behavioural Ecology of Frugivorous Tephritidae

Computational modelling of mechanisms underlying processes in the real world can be of great value in understanding complex biological behaviours. Uptake in general biology and ecology has been rapid. However, it often requires specific data sets that are overly costly in time and resources to collect. The aim of the current study was to test whether a generic behavioural ecology model constructed using published data could give realistic outputs for individual species. An individual-based model was developed using the Pattern-Oriented Modelling (POM) strategy and protocol, based on behavioural rules associated with insect movement choices. Frugivorous Tephritidae (fruit flies) were chosen because of economic significance in global agriculture and the multiple published data sets available for a range of species. The Queensland fruit fly (Qfly), Bactrocera tryoni, was identified as a suitable individual species for testing. Plant canopies with modified architecture were used to run predictive simulations. A field study was then conducted to validate our model predictions on how plant architecture affects fruit flies’ behaviours. Characteristics of plant architecture such as different shapes, e.g., closed-canopy and vase-shaped, affected fly movement patterns and time spent on host fruit. The number of visits to host fruit also differed between the edge and centre in closed-canopy plants. Compared to plant architecture, host fruit has less contribution to effects on flies’ movement patterns. The results from this model, combined with our field study and published empirical data suggest that placing fly traps in the upper canopy at the edge should work best. Such a modelling approach allows rapid testing of ideas about organismal interactions with environmental substrates in silico rather than in vivo, to generate new perspectives. Using published data provides a saving in time and resources. Adjustments for specific questions can be achieved by refinement of parameters based on targeted experiments.     

雄蝇追逐行为的分析

本文报告了在自由飞行条件下雄蝇追逐的行为实验及其分析的初步结果.其结果如下:1.追逐雄蝇水平方向偏转的角速度dF_1线性地依赖于目标蝇水平方位误差角T_1的大小.当目标在前视场中,即空间误差角|G|<π/4时,线性回归直线的斜率约为37**;而当空间误差角|G|>π/4时,线性回归直线的斜率约为6.7.2.追逐雄蝇俯仰方向偏转角速度dF_2在(-(π/2),π/2)的范围内线性依赖于俯仰误差角T_2的大小,其回归直线的斜率约为14.3.雄蝇追逐行为中,水平方位误差角频数分布的直方图呈现为峰值在零点的对称型分布;而俯仰误差角T_2频数分布的直方图是非对称型的,即仰角出现的频数大大超过俯角出现的频数.4.雄蝇主要利用了两蝇间距离变化dD的信息以及目标误差角来控制向前飞行的速度V.当误差角小时(即目标在前视场中),dD一般为负值,说明两蝇间的距离减小,而雄蝇追逐飞行的加速度A却与dD呈现正的线性关系.当误差角大时(即目标位于后视场中),dD一般为正值,说明两蝇间的距离增加.     

Gait posture estimation using wearable acceleration and gyro sensors

A method for gait analysis using wearable acceleration sensors and gyro sensors is proposed in this work. The volunteers wore sensor units that included a tri-axis acceleration sensor and three single axis gyro sensors. The angular velocity data measured by the gyro sensors were used to estimate the translational acceleration in the gait analysis. The translational acceleration was then subtracted from the acceleration sensor measurements to obtain the gravitational acceleration, giving the orientation of the lower limb segments. Segment orientation along with body measurements were used to obtain the positions of hip, knee, and ankle joints to create stick figure models of the volunteers. This method can measure the three-dimensional positions of joint centers of the hip, knee, and ankle during movement. Experiments were carried out on the normal gait of three healthy volunteers. As a result, the flexion–extension (F–E) and the adduction–abduction (A–A) joint angles of the hips and the flexion–extension (F–E) joint angles of the knees were calculated and compared with a camera motion capture system. The correlation coefficients were above 0.88 for the hip F–E, higher than 0.72 for the hip A–A, better than 0.92 for the knee F–E. A moving stick figure model of each volunteer was created to visually confirm the walking posture. Further, the knee and ankle joint trajectories in the horizontal plane showed that the left and right legs were bilaterally symmetric.     

Intraspecific variation in movement patterns: modeling individual behaviour in a large marine predator

In large marine predators, foraging entails movement. Quantitative models reveal how behaviours can mediate individual movement, such that deviations from a random pattern may reveal specific search tactics or behaviour. Using locations for 52 grey seals fitted with satellite-linked recorders on Sable Island; we modeled movement as a correlated random walk (CRW) for individual animals, at two temporal scales. Mean move length, turning angle, and net squared displacement (R²_n: the rate of change in area over time) at successive moves over 3 to 10 months were calculated. The distribution of move lengths of individual animals was compared to a Lévy distribution to determine if grey seals use a Lévy flight search tactic. Grey seals exhibited three types of movement as determined by CRW model fit: directed movers – animals displaying directed long distance travel that were significantly underpredicted by the CRW (23% of animals); residents – animals remaining in the area surrounding Sable Island that were overpredicted by the model (29% of animals); and correlated random walkers – those (48% of animals) in which movement was predicted by the CRW model. Kernel home range size differed significantly among all three movement types, as did travel speed, mean move length, mean R²_n and total distance traveled. Sex and season of deployment were significant predictors of movement type, with directed movers more likely to be male and residents more likely to be female. Only 30% of grey seals fit a Lévy distribution, which suggests that food patches used by the majority of seals are not randomly distributed. Intraspecific variation in movement behaviour is an important characteristic in grey seal foraging ecology, underscoring the need to account for such variability in developing models of habitat use and predation.     

On Higher Ground: How Well Can Dynamic Body Acceleration Determine Speed in Variable Terrain?

Introduction

Animal travel speed is an ecologically significant parameter, with implications for the study of energetics and animal behaviour. It is also necessary for the calculation of animal paths by dead-reckoning. Dead-reckoning uses heading and speed to calculate an animal’s path through its environment on a fine scale. It is often used in aquatic environments, where transmission telemetry is difficult. However, its adoption for tracking terrestrial animals is limited by our ability to measure speed accurately on a fine scale. Recently, tri-axial accelerometers have shown promise for estimating speed, but their accuracy appears affected by changes in substrate and surface gradients. The purpose of the present study was to evaluate four metrics of acceleration; Overall dynamic body acceleration (ODBA), vectorial dynamic body acceleration (VDBA), acceleration peak frequency and acceleration peak amplitude, as proxies for speed over hard, soft and inclined surfaces, using humans as a model species.

Results

A general linear model (GLM) showed a significant difference in the relationships between the metrics and speed depending on substrate or surface gradient. When the data from all surface types were considered together, VeDBA had the highest coefficient of determination.

Conclusions

All of the metrics showed some variation in their relationship with speed according to the surface type. This indicates that changes in the substrate or surface gradient during locomotion by animals would produce errors in speed estimates, and also in dead-reckoned tracks if they were calculated from speeds based entirely on a priori calibrations. However, we describe a method by which the relationship between acceleration metrics and speed can be corrected ad hoc, until tracks accord with periodic ground truthed positions, obtained via a secondary means (e.g. VHF or GPS telemetry). In this way, dead-reckoning provides a means to obtain fine scale movement data for terrestrial animals, without the need for additional data on substrate or gradient.     

Gait analysis using gravitational acceleration measured by wearable sensors

A novel method for measuring human gait posture using wearable sensor units is proposed. The sensor units consist of a tri-axial acceleration sensor and three gyro sensors aligned on three axes. The acceleration and angular velocity during walking were measured with seven sensor units worn on the abdomen and the lower limb segments (both thighs, shanks and feet). The three-dimensional positions of each joint are calculated from each segment length and joint angle. Joint angle can be estimated mechanically from the gravitational acceleration along the anterior axis of the segment. However, the acceleration data during walking includes three major components; translational acceleration, gravitational acceleration and external noise. Therefore, an optimization analysis was represented to separate only the gravitational acceleration from the acceleration data. Because the cyclic patterns of acceleration data can be found during constant walking, a FFT analysis was applied to obtain some characteristic frequencies in it. A pattern of gravitational acceleration was assumed using some parts of these characteristic frequencies. Every joint position was calculated from the pattern under the condition of physiological motion range of each joint. An optimized pattern of the gravitational acceleration was selected as a solution of an inverse problem. Gaits of three healthy volunteers were measured by walking for 20 s on a flat floor. As a result, the acceleration data of every segment was measured simultaneously. The characteristic three-dimensional walking could be shown by the expression using a stick figure model. In addition, the trajectories of the knee joint in the horizontal plane could be checked by visual imaging on a PC. Therefore, this method provides important quantitive information for gait diagnosis.     

Investigating behaviour and population dynamics of striped marlin (Kajikia audax) from the southwest Pacific Ocean with satellite tags

Behaviour and distribution of striped marlin within the southwest Pacific Ocean were investigated using electronic tagging data collected from 2005-2008. A continuous-time correlated random-walk Kalman filter was used to integrate double-tagging data exhibiting variable error structures into movement trajectories composed of regular time-steps. This state-space trajectory integration approach improved longitude and latitude error distributions by 38.5 km and 22.2 km respectively. Using these trajectories as inputs, a behavioural classification model was developed to infer when, and where, 'transiting' and 'area-restricted' (ARB) pseudo-behavioural states occurred. ARB tended to occur at shallower depths (108 ± 49 m) than did transiting behaviours (127 ± 57 m). A 16 day post-release period of diminished ARB activity suggests that patterns of behaviour were affected by the capture and/or tagging events, implying that tagged animals may exhibit atypical behaviour upon release. The striped marlin in this study dove deeper and spent greater time at ≥ 200 m depth than those in the central and eastern Pacific Ocean. As marlin reached tropical latitudes (20-21 °S) they consistently reversed directions, increased swimming speed and shifted to transiting behaviour. Reversals in the tropics also coincided with increases in swimming depth, including increased time ≥ 250 m. Our research provides enhanced understanding of the behavioural ecology of striped marlin. This has implications for the effectiveness of spatially explicit population models and we demonstrate the need to consider geographic variation when standardizing CPUE by depth, and provide data to inform natural and recreational fishing mortality parameters.     

Modelling larval movement data from individual bioassays

We consider modelling the movements of larvae using individual bioassays in which data are collected at a high‐frequency rate of five observations per second. The aim is to characterize the behaviour of the larvae when exposed to attractant and repellent compounds. Mixtures of diffusion processes, as well as Hidden Markov models, are proposed as models of larval movement. These models account for directed and localized movements, and successfully distinguish between the behaviour of larvae exposed to attractant and repellent compounds. A simulation study illustrates the advantage of using a Hidden Markov model rather than a simpler mixture model. Practical aspects of model estimation and inference are considered on extensive data collected in a study of novel approaches for the management of cabbage root fly.     

Decision tree classification of proteins identified by mass spectrometry of blood serum samples from people with and without lung cancer

A classification and regression tree (CART) model was trained to classify 41 clinical specimens as disease/nondisease based on 26 variables computed from the mass-to-charge ratio (m/z) and peak heights of proteins identified by mass spectroscopy. The CART model built on all of the specimens (no cross-validation) had an error rate of 4/41 = 10%. The CART model suggests that mass spectra peaks in the 8000-10,000, 20,000-30,000, 45,000-60, 000, and >125,000 m/z ranges may be valuable in distinguishing between the disease/nondisease specimens. The area under the receiver operating characteristics curve was 0.80 +/- 0.07 for leave-one-out cross-validation.