Prediction model for daily feed intake during the growing period in floor-reared Wanxi White Geese based on machine learning and feeding behavior features

To improve the efficiency of obtaining daily feed intake (DFI) of Wanxi White geese under floor-rearing conditions, this study used 200 Wanxi White geese as experimental animals. Individual information such as body weight (BW) and feeding behavior data were collected using Radio Frequency Identification (RFID) technology and an electronic feeder system, and feature indicators were constructed on a daily basis to form four feature sets. Four machine learning (ML) algorithms were used to develop 16 DFI prediction models by combining different feature sets, and model performance was evaluated across three weekly age segments during the growing period (22-28 d, 29-35 d, and 36-42 d). The results showed that ensemble learning models generally outperformed the linear model, and the integrated feature set combining individual information and feeding behavior features exhibited the most stable performance. The RF-IF_FB model performed best in the 22-28 d stage (R² = 0.75, RMSE = 13.96 g), the GBR-IF_FB model performed best in the 29-35 d stage (R² = 0.77, RMSE = 14.65 g), and the XGB-IF_FB model performed best in the 36-42 d stage (R² = 0.77, RMSE = 14.82 g). Furthermore, weekly feed conversion ratio (FCR) was estimated based on predicted DFI, and geese were classified according to the mean and standard deviation of FCR. The predicted grades showed high agreement with the measured grades (kappa = 0.74-0.80). These findings indicate that ML and feeding behavior features can be used to predict DFI of Wanxi White geese under floor-rearing conditions and support efficiency grading based on FCR derived from predicted values.