This dataset comprises multi-modal behavioural and accelerometry data collected from four captive platypuses (Ornithorhynchus anatinus). It includes raw tri-axial acceleration time-series data (±8 g, 50 Hz), annotated behavioural segments derived from synchronised video observations, and derived predictor variables used for behavioural classification modelling. Behavioural annotations are provided as time-aligned windows corresponding to five classes (burrow rest, surface rest, grooming, foraging, and diving), with window lengths tailored to behavioural duration.

Processed datasets include raw axis accelerometer data, dynamic and statics acceleration signals for each axis at the window level, alongside model-ready datasets with summarised accelerometer variables used for machine learning analyses. Accompanying R scripts provide full reproducibility of data processing, feature extraction, and Random Forest model implementation.

Video recordings used for behavioural validation are not publicly available due to sensitivity and ethical restrictions but may be accessed upon reasonable request subject to institutional approval. Metadata describing sampling protocols, sensor specifications, data structure, and variable definitions are included to facilitate reuse and interpretation.