An Explainable AI Framework Combining Multi- Task GCN and Reinforcement Learning for Adaptive Feeding and Animal Health Management in Dairy Farming

Interweaving simulation modelling with XAI based
frameworks like Multi-Task Graph Convolutional Networks
(Free-Range-Learning, where Deep Reinforcement Learning
shown here using Genetic Algorithms &SHapley Additive
exPlanations (SHAP) to improve decision-making to produce
XAI to dairy farm management. Despite the difficulty inherent
to such a large dataset of over 10,000 animal-days of
multimodal sensor data input, the developed framework
attained highly impressive predictive performance including a
classification accuracy of 92.3% along with 90.1% precision
and 90.5% recall. For the regression tasks MSE of 0.038 and R2
of 0.87 were obtained. Furthermore, the DRL component,
trained using Proximal Policy Optimization (PPO), was able to
make a 25% improvement in feed efficiency and 12.4%
decrease in cost of feeding. by/via GAs led to a 9.7 percent
increase in sustainability score. SHAP-based interpretability
enabled an average user trust score of 4.3/5 and interpretability
score of 88.5% accuracy, demonstrating both the transparency
and applicability of this framework to the real world.
Collectively, these results highlight the framework’s potential
for providing high-quality, low-cost, interpretable solutions
suitable for adoption in precision dairy farming paradigms.