Cow behavior classification: The optimal set of parameters for the random Forest algorithm

Abstract

Accelerometer data are key in animal behavior classification systems using pet-mounted accelerometers. Behavioral data reflecting the health status, early detection of some diseases of cows, monitoring the health of cows through behavior are effective support tools for ranchers to help improve performance and save money cost. In the previous study, we proposed feature sets, data windows and used a random forest algorithm to classify four important cow behaviors, including: eating, lying, standing and walking. In this study, in order to improve the performance of the classification system, we propose to use suitable values ​​for the important parameter set of the random forest algorithm on the experimental data set. The experimental results show that with the value of the parameter set: number of trees = 25 and depth = 15, the classification performance is good with an accuracy of 95,9%.