Predicting Next Service Actions in Outpatient Care Using Sequential Operational Data

Outpatient care pathways are inherently fragmented, noisy, and weakly structured, making systematic modeling and prediction challenging. While extensive research has focusedon inpatient trajectories, outpatient service sequences remain largely underexplored, particularly using real-world operational data.In this study, we analyze outpatient care episodes from a real clinical center and formulatethe task of predicting the next service action based on a short history of prior services. Rawservice logs were cleaned, normalized, and transformed into sequential episodes. Serviceprefixes were embedded using a pretrained language model and classified using both linearand gradient-boosted machine learning models.Despite the absence of diagnostic labels, our results demonstrate that short-term servicehistory contains meaningful predictive signal, especially for visit-related actions. The findings establish a realistic baseline for next-action prediction in outpatient care and highlightdirections for data-enriched and clinically informed future models.