Forecasting egg production in free-range laying hens using multi-farm data

Despite technological advancements in poultry production, accurately forecasting laying rates including sudden drops in egg production remains challenging. This study evaluated the potential of predictive modelling approaches to forecast laying rates and detect reduced production days using data integrated from multiple commercial free-range hen farms. Historical production and weather data from four commercial free-range farms (Farms A-D), comprising 106 flocks and 35,346 flock-days, were analysed. Three Random Forest models were developed: (1) a single-farm model using data from Farm A, (2) a multi-farm model integrating data from Farms B, C, and D, and (3) a combined model incorporating all farms (A-D). Model performance was assessed using Farm A as the target farm. Predictive accuracy was evaluated using the Area Under the Receiver Operating Characteristic Curve (AUC) for classification and Root Mean Squared Error (RMSE) for regression tasks. The single-farm model achieved a median AUC of 0.86 and an RMSE of 2.8, demonstrating strong predictive ability. The multi-farm model achieved a slightly higher AUC (0.89) but lower regression accuracy (RMSE = 4.98). The combined model produced mixed outcomes, improving regression performance (RMSE = 2.55) but resulting in the lowest accuracy (AUC = 0.82). These results demonstrate that models developed using data from one farm can be effectively applied to another, highlighting the potential for cross-farm prediction. Overall, the findings suggest that integrating data from multiple farms can support forecasting of egg production in free-range systems, although combining datasets does not consistently improve model performance. This approach provides a basis for developing practical tools to assist producers in anticipating production changes and improving farm management.