Application note: Validation of BovHEAT — An open-source analysis tool to process data from automated activity monitoring systems in dairy cattle for estrus detection

Abstract

This application note introduces the software called Bovine Heat Detection and Analysis Tool (BovHEAT), a validated and open-source analysis tool to process automated activity monitoring (AAM) data for estrus detection. We used activity data collected from a neck-attached accelerometer (Heatime, SCR Engineers Ltd., Netanya, Israel) that is widely adopted in the dairy industry. Developed with the Python programming language, BovHEAT offers fully automatic and scalable processing for estrus detection with additional functionality for handling missing values and a plausibility check for timing of events. Processed output is provided in an Excel file with result tables in the long and wide format. Additionally, a PDF file containing activity change line graphs is generated. For validation, we compared the accuracy and time of three different methods to process AAM data: 1) manual data evaluation (MAN), 2) Excel tool (EXCEL), and 3) BovHEAT. Two different datasets from 8 farms (1 farm in Canada; 7 farms in Germany) were used. Validation was performed independently by three investigators. In total, activity data from 60 cows representing a maximum number of 600 observations (50 days with 12 observations per day) per cow were used. Manual data evaluation was less accurate due to transcription errors, with 13 of 60 cows having at least one error. More specifically, 16 out of 110 estrus events were recorded incorrectly. The time to process AAM data and transfer the results into a standardized results table for 10 cows was 41.0 (range 28–53) minutes, 30.7 (18–48) minutes, and 11.7 (4–16) minutes for MAN, EXCEL and BovHEAT, respectively. Without the standardized results table, a fully automated run with BovHEAT processing the complete dataset of 5,477 cows, which consisted of 361 XLS and XLSX files, took 172 s. The results from this study indicate that BovHEAT speeds up processing, requires less user interaction and provides additional features. Our aim is to accelerate future research with AAM data and facilitate reproducibility via our validated analysis tool. Since BovHEAT is open-source and MIT-licensed, it allows customization to support different sensors and manufacturers. The BovHEAT tool can be evaluated, downloaded and contributed to on GitHub (https://github.com/bovheat/bovheat, https://doi.org/10.5281/zenodo.3890126).