108 Potential of Accelerometers to Remotely Detect Bovine Ephemeral Fever in Heifers Using Pattern Recognition

Abstract

Abstract Bovine Ephemeral Fever (BEF), an arthropod-borne rhabdovirus, is widespread in tropical and subtropical rangelands. It affects cattle with symptoms of fever, lameness, and inappetence, and can be a major economic impact. The goal of this study was to determine if accelerometer data can be used to identify the behavior patterns that occur when heifers become ill from BEF. The study was conducted from 24 August to 2 October 2016 in Rockhampton, Queensland, Australia. Eight heifers were monitored with 3-axis accelerometer sensors attached to collars. During the study, two heifers (1,402 and 1,413) were observed with BEF by the manager on 12 September and 1 October 2016, respectively. In this study, we demonstrated the potential of accelerometer data to autonomously detect the pattern and recognize when the heifers become sick by applying two different approaches, cosine similarity (CS) and deviation from previous behavioral patterns. Movement intensity (MI) and movement variation (MV) were calculated from accelerometer data using 1-minute epochs and then averaged into hour periods. Cosine similarity is a cosine of the angle between two data instances (MI or MV) that can be used to quantify their similarity. The time of illness detection by a veterinarian was compared with CS between the day of prediction and the average of 3 days before the prediction. Deviation is a measurement to calculate the sum or sum squares of the absolute difference of MI or MV between a day of prediction and the mean of 3 days before prediction for each corresponding hour. This approach is based on a hypothesis that when a heifer is sick, the heifer is likely less active during times it is normally active. When the heifer became ill with BEF, there were no similarities between the day of prediction and its 3 previous days resulting in a small CS value and potential illness detection, while there was a large difference between the day of prediction and its 3 previous days resulting in a high deviation value and a potential BEF detection metric. To determine how large deviation is or how small CS is to be considered abnormal behavior, the concept of outliers was evaluated using boxplots. A CS or deviation is an outlier if it is distant from other values of an individual animal. Our approaches show that heifer 1402 had behavioral changes one day before the manager observed BEF, and heifer 1413 had behavioral changes on the same day the manager observed BEF. This case study demonstrates the potential of using accelerometer data to detect disease autonomously. However, more research is to minimize false positives that may occur from other similar diseases, abnormal weather events or cyclical changes in behavior such as ostrus.