Imitation learning for dynamic home healthcare routing and scheduling with stochastic patients

We study a stochastic and dynamic home healthcare routing and scheduling problem, where a service provider must decide daily on new patient acceptance, visit days allocation, and caregiver routing under uncertain patient arrival and length of stay. To address this challenge, we develop a hybrid framework integrating machine learning and combinatorial optimization. A neural network trained in an imitation learning fashion captures insights from perfect information solutions in order to guide daily decision-making. We provide preliminary results for instances that mimic real-life systems.