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

Abstract

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.

Fig. 1

a Examples of the five behaviours for classification. Overall dynamic body acceleration (ODBA) is calculated as the sum of the absolute values of dynamic acceleration from the three axes. b Dynamic acceleration in the three orthogonal axes: sway (blue), heave (red) and surge (grey) during each behaviour and c corresponding wavelet spectrum generated from the sway axis showing increased signal strength amplitude during the burst and headshake event

Fig. 2

An example from the sway acceleration axis of a 63 s prey manipulation event, consisting of three headshakes (HS; totalling 19 s) and a brief burst event

Fig. 3

Estimated smoother for the effect of hour of day on the probability of headshaking behaviour occurring by the juvenile lemon shark in Bimini, Bahamas. The lowest and highest probabilities of a headshake occurring are around 0800 and 1700 h, respectively. Estimates are based on final binomial generalized additive mixed model. The solid line is the smoother. Dark grey shaded area surrounding the smoother represent 95% confidence intervals. The light grey shaded area represents the range of sunset times throughout the deployments. The dashed line represents the mean likelihood of a headshaking occurring. The blue dots represent mean hourly temperature (°C), calculated from the temperature sensor in the acceleration data logger (ADL) packages, across all deployments