Classification of dairy cow excretory events using a tail-mounted accelerometer

Grazing livestock contributes to pasture nitrogen (N) through urine and faeces N losses in pasture-based systems are recognized as an important consideration for sustainable land management. Knowing the frequency quantity of excreta produced dairy production could be useful informing best management practice. The aim this experiment was determine whether data from tail-mounted accelerometers used classify cow excretory events. Ten non-lactating Holstein cows were fitted with a accelerometer set record at 1 Hz individually observed 3.3–5.3 h each. recorded behaviours urination, defecation, standing lying (both left right laterality). G-force acceleration values X, Y Z axes downloaded windows varying sizes (3-, 6-, 9-, 12- 15 s) extract basic features (mean, minimum, maximum SD) consecutive sequences each behaviour. Windows stepped forward by s feature extraction five datasets developed. Data all compiled random forest algorithm model development. times 10-fold stratified cross-validation (SCV) evaluate window sizes. Sensitivity precision always exceeded 84% postures both unbalanced balanced datasets. Classification performance events improved significantly (P < 0.01) size increased. Due performance, selected further tests full set. Random models developed using leave-one-cow-out strategy (model n-1 evaluated on held-out cow). remained high (sensitivity & ≥ 91 %) but poor highly variable. SCV results clearly optimistic more needed testing. However, it may also necessary develop test individual animal comparison because there considerable variation between animals