Hospital capacity management depends on accurate real-time estimates of hospital-wide discharges. Estimation of patient discharge by a clinician requires a large amount of effort and, even when attempted, accuracy in forecasting next-day patient-level discharge is poor. A machine learning approach may support next-day discharge predictions by incorporating electronic health record (EHR) audit log data for discharge prediction. Such an approach can assist hospital administrators in more accurately predicting time of discharge, with the potential of aligning timely care services with a patient’s needs and streamlining inpatient flow of hospitals.