Forecasting egg production performance and fluctuations in commercial free-range poultry systems using a random forest model

• Variability in egg production presents significant challenges for free-range poultry systems, impacting productivity, profitability, and resource management. • Random Forest models were applied to both regression (forecasting laying rates) and classification (predicting days with low egg production) tasks using aggregated data features from 7-, 14-, 21-, and 28-day moving windows. • Classification tasks achieved high performance with AUC values above 0.9 and sensitivity scores exceeding 0.85, effectively identifying low production days. • Regression tasks demonstrated RMSE values around 2.5%, indicating precise laying rate predictions with lower error rates than prior studies. The free-range poultry industry is faced with numerous challenges that contribute significantly to the flock’s variability in egg production. Forecasting fluctuations and egg laying rate for commercial flocks is important as it allows early implementation of proactive farm management decisions thereby minimising unexpected interruptions to the production rate. This study employed a Random Forest model to forecast egg production fluctuations and near-future laying rates of commercial free-range hens. Datasets from a single free-range commercial farm, comprising 7 flocks including production and environmental variables were used in a machine learning workflow. The workflow involved the use of a classification task to detect problematic fluctuations in egg production and a regression task to forecast laying rates. This approach provides an understanding of the requirements to forecast production measures feasibly with a level of sensitivity and precision suitable for a decision support system. The results from this study showed that the 28-day data window had the best performance, with a 5-day forecast interval. For the classification task, the AUC values were above 0.9 and sensitivity scores exceeded 0.85 indicating the model's ability to predict the problematic production days, while PPV values around 0.4 suggests a relatively high rate of false positives. For the regression task, the RMSE value was 2.5% demonstrating accurate forecasting of laying rates, with lower error rates. Feature importance analysis revealed that production variables such as laying and mortality rates strongly predict laying performance rather than environmental variables. The findings from this study will build towards the development a decision support system for free-range egg producers.