A High-Recall Ensemble Framework for Patient No-Show Prediction; Risk-Adjusted Scheduling Optimization in Healthcare Systems

Abstract

B Patient no-shows significantly disrupt healthcare delivery, leading to underutilized resources and financial strain. This study proposes a hybrid predictive–prescriptive framework that integrates machine learning with a Reinforcement Learning (RL)-inspired decision engine for clinical queue optimization. We have created a model consisting of an ensemble of XGBoost with CatBoost that are specifically tuned to high-risk detection with Recall of 87.75% and F1-score of 0.429. Unlike traditional models that prioritize raw accuracy, our approach focuses on minimizing missed no-show cases to drive a three-tier intervention strategy: automated SMS reminders, targeted telephonic confirmations, and risk-adjusted smart overbooking. The predicted probabilities are integrated into an
RL-based scheduling mechanism that dynamically categorizes appointments into "Confirm" or "Reschedule" actions based on a probability-cost trade-off. Experimental results demonstrate that this framework transforms predictive insights into actionable scheduling decisions, effectively reducing idle slots and maximizing clinical throughput. The proposed system can provide a scalable, practical solution to the problem of improving the delivery of healthcare services and operational efficiency in the resource-constrained settings by changing the current system of static scheduling to that of probability-based intervention