Intelligent classification of swine behaviors using an inertial sensor: Effects of data features and Machine Learning models on classification performance

Wearable electronics and machine learning (ML) have enabled the synchronous recording and analysis of animal motion data, offering new opportunities for precision livestock monitoring. While swine behavior classification using multidimensional sensor data has gained attention, existing approaches often lack systematic comparisons of feature importance and model performance across diverse behaviors. The purpose of this study is to employ an inertial measurement unit (IMU) data to classify swine behaviors, compare time- and frequency-domain features, and evaluate their classification performance using multiple ML algorithms. The impact of individual features and model selection on prediction scores is quantified, identifying key discriminators for each behavior. This study provides a data-driven framework for high-accuracy swine behavior classification, demonstrating that feature selection and model choice significantly affect performance, advancing the development of robust, ML-enabled tools for smart agricultural management.