Inclusion of features derived from a mixture of time window sizes improved classification accuracy of machine learning algorithms for sheep grazing behaviours

What if farmers could tell if one of their livestock was sick, without even being around the animal? Farmers are now using inertial motion sensors to study grazing behaviour. But the challenge is interpreting all the data those sensors provide. New research led by ACEMS PhD candidate Shuwen Hu reveals which machine learning methods are best suited to handle the problem.

The project is a collaboration with CSIRO and it looked specifically at sheep. In her study, she used data collected by CSIRO from accelerometers attached to the neck of 17 Merino sheep over a period of two days. CSIRO also provided a ground truth dataset of behaviour recordings (grazing, ruminating, walking, and standing) over the same time period. Shuwen then investigated the ability of three machine learning (ML) approaches, Random Forest (RF), Support Vector Machine (SVM) and linear discriminant analysis (LDA), to accurately classify sheep behaviour. Her results clearly showed that simultaneous inclusion of features derived from time windows of mixed sizes, ranging from 2 seconds to 15 seconds, significantly improved the behaviour classification accuracy, in comparison to those determined from a single unique time window size. Of the three ML methods applied here, the RF approach yielded the best results.  

“Developing a new combined window size approach enabled us not only to minimise subjective selection of window sizes, but also to systematically and simultaneously consider the features from multiple window sizes together to capture the irregular duration of animal behaviours, that occur both within and across behaviours,” says Shuwen.

The research is available in Computers and Electronics in Agriculture.