Development of a Machine Learning-Based Framework for Improving Appointment Attendance Prediction in Outpatient Clinics Using Stacking Algorithm

Abstract

Patient no-shows pose a significant challenge for healthcare providers, impacting operational efficiency and resource allocation. This study introduces an innovative machine learning framework designed to predict patient no-shows in outpatient clinics, while addressing data imbalance challenges in healthcare datasets and identifying crucial factors that influence patient attendance. The methodology encompasses data preprocessing, recursive feature elimination for feature selection, novel feature extraction techniques, and the utilization of methods such as the synthetic minority over-sampling technique and stratified k-fold cross-validation to handle class imbalance. The results highlighted the efficacy of a stacking model that integrates six distinct algorithms, achieving an impressive area under the receiver operating characteristic curve (AUC-ROC) of 0.911, an accuracy of 82%, a recall of 82%, and a precision of 85%. Finally, the technique for order of preference by similarity to ideal solution was employed based on evaluation metrics, demonstrating the superiority of the stacking model compared with other classifiers.