Efficient Estimation of Cow's Location Using Machine Learning Based on Sensor Data

Indoor localization of dairy cows is important for determining cow behavior and enabling an effective farm management. In this study, a low-cost localization system was constructed by attaching accelerometers to dairy cows kept indoors in a barn in order to obtain radio wave strength. Using link quality indicator (LQI) data, we employed four machine learning models to predict the position of the cow: LightGBM, logistic regression, support vector machine (SVM), and neural network. The prediction performance and computational cost of the models were compared and evaluated. In the monitoring and building of the prediction models for cow's location, we considered various sizes of location (barn) compartments and evaluated the performance of each prediction model using with different compartments. The experimental results showed that LightGBM and neural networks have an accuracy of 46.6% at 9m horizontal and 12m vertical and an accuracy of 90% at 45m horizontal and 15m vertical. In terms of the computational score, we may consider whether to use neural network or LightGBM depending on the amount of data to be predicted at a time in the location estimation system.