Event-Based Dual-Task Forecasting for SLA-Oriented Hospital Transport Operations Using Machine and Deep Learning Models

Service Level Agreement (SLA) compliance in hospital transport processes is essential in terms of patient safety, service continuity, and resource efficiency. However, transport requests occur as irregular events, limiting the applicability of equally spaced time-series assumptions. The presented study jointly addresses two complementary objectives in an event-based framework: predicting the interarrival time between consecutive transport requests (next-event forecasting) and forecasting the total request count within forward SLA horizons (forward-count forecasting). Machine learning methods such as Ridge Regression, Extra Trees, and Histogram-based Gradient Boosting, as well as deep learning architectures such as Long Short-Term Memory and Gated Recurrent Unit, were compared under different time horizons and adaptive history windows on time-stamped transport request records from the operational system supporting a private hospital in Turkey, including patient, specimen, and material transport requests. Results indicate that deep learning methods yield lower errors in demand count prediction at short time horizons; as the horizon lengthens, machine learning performs similarly and even outperforms in some cases; and as the history window increases, the prediction error for the next request occurrence systematically decreases. The lowest mean absolute error values in request counts were obtained for demand forecasting within a 30 min time window; 2.10 for material transport, 3.88 for patient transport, and 2.84 for specimen transport. Additionally, R2 value reached 0.98 for next-event forecasting with a rolling-memory window of 20 events. Overall, the findings suggest that hospital transport demand is substantially predictable and that event-based forecasting can support SLA-oriented staffing, task dispatching, and delay mitigation.