Classifying animal behavior from accelerometry data via recurrent neural networks

Liang Wang, Reza Arablouei, Flavio A. P. Alvarenga, Gregory Bishop-Hurley

Abstract: Highlights•We develop RNN-based models for animal behavior classification via IMU sensor data.•We utilize tri-axial accelerometry data collected using cattle collar and ear tags.•We evaluate the accuracy and complexity of various LSTM- and GRU-based models.•RNN-based models can be more accurate and less complex than the CNN-based models.•GRU-based models outdo LSTM-based ones and offer good accuracy-complexity tradeoff.