Performance analysis of machine learning classifier on different cow health datasets

Animal welfare and good health are crucial for dairy cow farms and long-term milk production. Unfortunately, daily animal condition monitoring is challenging, especially on big farms where staff members need more time to observe animals and spot early signs of sickness. The machine learning algorithm can be constructive in continuously monitoring animal conditions. The performance of different classifiers is studied here, based on the CDAC-developed innovative farm-based cow health monitoring system. The system comprises hardware components, a cloud infrastructure, a client application, and cutting-edge data measurement and analysis algorithms. The device can successfully monitor animal well-being and the estrus cycle after being tested in a real-world setting. In this study, data was collected from the IMU sensor attached to the neck of the cows. Along with this manual data annotation, CCTV footage of the livestock was also recorded as a means of data verification. The help of machine learning algorithms, Random Forest (RF), XGBoost (XGB), and Support Vector Machine (SVM) to classify the behavior of the cattle were taken. The algorithms were used on datasets with different features, showing additional accuracy with the other featured datasets. A brief study was presented about the improvement and change in accuracy after using the machine learning algorithm on different datasets with different features.