Classifying horse activities with big data using machine learning

Abstract

Using big data-assisted machine learning methods in animal science has received increasing attention in recent years since they extract useful insights from large-scale animal datasets. Especially, animal activity recognition is the task of identifying the actions performed by animals and can provide rich insight into their health, welfare, reproduction, survival, foraging, and interaction with humans/other animals. This paper aims to propose a new solution for this purpose by building a machine learning model that classifies the actions of horses based on big sensor data. Unlike the previous studies, our study is original in that it compares the accuracies of per-subject (personalized) and cross-subject (generalized) models. It is the first study that especially compares different ensemble learning algorithms for horse activity recognition in terms of classification accuracy, including bagging trees, extremely randomized trees, random forest, extreme gradient boosting, light gradient boosting, gradient boosting, and categorical boosting. The purpose of the study is to classify five horse activities: walking, standing, grazing, galloping, and trotting. The experimental results showed that our solution achieved very good performance (94.62%) on average on a real-world dataset. Furthermore, the results also showed that our method outperformed the state-of-the-art methods on the same dataset.