Deep Learning–Assisted Appointment Scheduling Under Uncertainty

This paper studies the appointment scheduling problem where heterogeneous appointments (jobs) are processed by multiple providers (machines). Our goal is to optimize provider utilization and appointment wait time under uncertainties (processing time, punctuality, and no-show). We propose a learning-based algorithm combining deep learning and partial mixed integer programming. Specifically, we develop a convolutional neural network that captures uncertainties and estimates associated costs. It is trained on synthetically generated data and then integrated into the deterministic model’s objective. Under nondeterministic conditions, our approach performs well, showing an average improvement of over 157% on the largest instance compared with a heuristic. Through simulation, we compare deterministic and nondeterministic solutions: deterministic solutions yield about 16% higher utilization rewards, whereas nondeterministic solutions reduce wait time and overtime penalties by around 80% and 82%, respectively (99% confidence). Our results also show that a more flexible unpunctuality management policy reduces appointment wait time and machine overtime but increases job tardiness because it often necessitates serving jobs out of order, potentially encouraging further unpunctuality. History: Yu Ding served as the senior editor for this article. Funding: Financial support from the Natural Sciences and Engineering Research Council of Canada (NSERC), [Grant RGPIN-2020-04301]. Data Ethics & Reproducibility Note: The appendix and code capsule are available in the e-Companion to this article (available at https://doi.org/10.1287/ijds.2025.0090 ).