PSXI-22 Prediction quality of cattle behavior traits evaluated through different cross-validation strategies using wearable sensor data and machine learning algorithms

Abstract

Abstract Wearable sensors have been adopted as an alternative for real-time monitoring of cattle feeding behavior in grazing systems. However, even using machine learning (ML) techniques confounding effects such as cross-validation strategy may inflate the prediction quality. Our objective was to evaluate the effect of different cross-validation strategies on the prediction of grazing activities in cattle using wearable sensor data and ML algorithms. Six Nellore bulls (345 ± 21 kg) had their behavior visually classified as grazing or not-grazing for a period of 15 days. Generalized Linear Model (GLM), Random Forest (RF), and Artificial Neural Network (ANN) were employed to predict behavior (grazing or not-grazing) using 3-axis accelerometer data. For each analytical method, three cross-validation strategies were evaluated: holdout, leave-one-animal-out (LOAO), and leave-one-day-out (LODO). Algorithms were trained using similar dataset sizes (holdout: n = 57,862; LOAO: n = 56,786; LODO: n = 56,672). Regardless of the cross-validation strategy, GLM achieved the worst prediction accuracy (53%) compared to the ML techniques (65% for both RF and ANN). ANN performed slightly better than RF for LOAO (73%) and LODO (64%) cross-validation strategies. The holdout yielded the highest accuracy values for all three ML approaches (GLM: 59%, RF: 76%, and ANN: 74%), followed by LODO (58%) and LOAO (55%). In conclusion, the GLM approach was not adequate to predict grazing behavior, regardless of the cross-validation strategy. The greater prediction accuracy observed for holdout cross-validation may simply indicate a lack of data independence and the presence of carry-over effects from animals and grazing management. Our results suggest that generalizing predictive models to unknown (not used for training) animals or grazing management may incur in poor prediction quality. The results highlight the need for using biological knowledge to define the validation strategy that is closer to the real-life situation.