Understanding the behaviours of free-ranging animals over biologically meaningful time scales (e.g. diel, tidal, lunar, seasons) gives important insights into their ecology. Bio-logging tools allow the remote study of elusive animals by recording high resolution movement data. Machine learning (ML) is common for automatic classification of behaviour from large data sets. To present a frame work for programming bio-loggers, the impact of sampling frequency on classification of behaviours was assessed.

Behavioural ethograms (swim, rest, chafe, burst and headshake) were developed for juvenile lemon sharks (Negaprion brevirostris) by observing sharks equipped with accelerometers during semi-captive trials at Bimini, Bahamas. Observations were used to ground truth data for ML. A random forest (RF) algorithm was developed to test a range of sampling frequencies.

Best overall classification was achieved at 30 Hz; however 5 Hz was appropriate for classification of swim and rest. Behaviours characterised by complex movements (headshake, burst, chafe) were not classified as well as swim and rest; classifier performance was best at 30 Hz.

Implications for classification of shark behaviour are discussed. These findings enabling us to refine classification of shark behaviour from bio-logging tools in the future.