Deep Learning performance in predicting dairy cows’ behaviour from a tri-axial accelerometer data

Abstract The accurate detection of behavioural changes represents a promising method to early reveal the onset of diseases in dairy cows. This study assessed the performance of deep learning (DL) in classifying dairy cows’ behaviour from accelerometery data and compared the results with those of classical machine learning (ML). Twelve cows with a tri-axial accelerometer were observed for 136 ± 29 min each to detect 5 main behaviours. For each 8s time-interval 15 metrics were calculated obtaining a dataset of 211,720 observation units and 15 columns. The whole dataset was randomly split into training (80%) and testing (20%). An 8-layer Convolutional Neural Network (CNN) was made of 3 convolution, 1 dropout, 1 max-pooling, 1 flattening and 2 dense layers. The CNN accuracy, precision and sensitivity/recall were calculated and compared with the performance of classical ML. The CNN overall accuracy and F1-score were equal to 0.94. The precision, sensitivity/recall, and F1-score of single behaviours had the following ranges 0.88–0.99, 0.88–0.99 and 0.89–0.99, respectively. The CNN outperformed all classical ML algorithms. The CNN in our specific raising conditions showed an overall high performance in successfully predicting multiple behaviours using a single accelerometer. Further studies considering different breeds, housing conditions and sensors are warranted.