Transparency and Fairness in Arrival Prediction

This project predicts patient arrivals at the Emergency Department (ED) of a Hospital using machine learning and probabilistic modeling. The dataset includes arrival times, visit start times, priority codes, and outcomes. A neuro-probabilistic model is developed with TensorFlow Probability, combining neural networks and probabilistic layers to pre- dict arrival rates. The model optimizes Poisson likelihood with custom loss functions, using features such as hour and weekday. Results show high predictive accuracy and the ability to compute confidence intervals, enhancing forecast reliability. This hybrid model improves healthcare de- mand forecasting and ED operational efficiency.