Hierarchical Long Short Term Memory Recurrent Neural Network for Goats Behaviour Prediction from Accelerometer Data

International audience

anglais. Gastrointestinal parasitism is a major challenge in small grazing ruminants, affecting animal welfare and farmers’income. In this regard, monitoring individual animal behaviour could help to develop new selection schemesthat favor animals with a lower risk of larval infestation, but also support the targeted use of anthelmintic byfocusing only on infested animals. Accelerometer sensors are widely used in combination with statistical modelsto predict the behaviour of grazing ruminants, but the lack of generalization of the models and the limited rangeof well-predicted behaviours are still challenging. In our study, we introduce an innovative methodology based onhierarchical long short term memory (LSTM) recurrent neural networks to predict the main behaviours of goatson pasture. For that purpose, we collected accelerometer data from the horns of 59 Creole goats and annotatedthe behaviour over 144 hours of data. We defined 13 moving features that are mathematical combinations ofthe raw data to get more information while preserving the temporal structure of the accelerometer time-series. Adata augmentation technique involving the addition of random noise was applied to sequences from the minoritybehaviour labels. A hierarchical LSTM model was then built to derive behaviours from a given accelerometersignal, by sequentially combining several models that first tackle simple classification tasks (e.g., grazing ornon-grazing segments), then increasingly complex ones (e.g., displacement or other activities), progressivelywithdrawing segments that have already been identified. The hierarchical LSTM model was validated using atesting set consisting of goats not seen during model training, and carefully selected to maximize behaviourallabels heterogeneity. Performance of the hierarchical LSTM model was also compared to those of a regular LSTMmodel which directly classified the raw signal into the 5 behaviours, used as the baseline. Highest performancewas obtained with the hierarchical LSTM model, reaching a Fscore of 87.84%, a precision of 89.44% and a recallof 86.3%. Best performance was obtained for grazing prediction (recall: 99.5%; precision: 99.4%), followingby resting (recall: 98.0%; precision: 98.4%) and ruminating (recall: 95.2%; precision: 89%). Most confusionsoccurred with the displacement (recall: 51.2%; precision: 67.8%), likely due to the low number of sequencesin the dataset (0.42% of the dataset). While other avenues remain to be explored for improving the predictionof such rare behaviours, our approach introduces key innovations that not only address the methodologicallimitations identified in the literature, but also facilitate further exploration of the role of goat behaviour inmanaging gastrointestinal parasitism on pasture.