Engineering an AI-enabled mobile expert decision-support system for integrated poultry farm operations

Abstract

Poultry farm management spans all the necessary operations needed to keep the business both sustainable and profitable. This domain however is plagued with challenges often beyond the control of most users who do not have formal knowledge of the domain. This research, in contrast to the data-driven machine learning methods developed and validated AI-enabled mobile expert decision support system that integrates experts-driven models for routine feeding, treatment, biosecurity and notifications. The models were developed based on both historical data from textbooks and internet, and interviews from domain experts. A 56 days simulation of 50 broilers under a stocking density of approximately 5.56  and 8.33  respectively. Regression metrics were used for evaluating RFM and RTM, while classification metrics were used for RBM and RNM. Results showed strong performance of the RFM (MAE = 218.9 g, RMSE = 287.7 g, MAPE = 5.3%, R² = 0.97, with 83.9% predictions within ±10% tolerance). The RTM achieved MAE = 3.4 units, RMSE = 5.5 units, MAPE = 6.9%, R² = 0.94, with 67.9% within tolerance. RBM and RNM attained high recall values with lower false alarm rates. The introduction of half-noise validation further improved accuracy and pushed tolerance compliance above 90%. These findings confirm the potentials of this integrated system in democratizing poultry management for smallholder farmers with no formal education.