A Classification of Hospitals using Performance Features and Machine Learning Algorithms in the Event of Random Surge in Inpatients and Prediction of Live Discharges

Abstract

In this paper, in phase-1 we classified hospitals of three categories in the presence of seven performance features, using several MATLAB^® machine learning algorithms that comprised ‘Classification learners’, and compared model validation and classification results. We chose the best trained model validated for prediction of future hospital category, while comparing the models based on accuracy rate and AUC (area under ROC curve). Out of the five classifiers that we short listed for final selection, Ensemble Subspace KNN (k-nearest neighbor) emerged as the best classifier, and it predicted the same original hospitals for treatment in the event of a random surge in Inpatients during a pandemic, indicating that those hospitals can handle the surge with the available resources. Then in phase-2 with the new Inpatients and predicted hospitals, we used ‘Regression learners’ and selected the best learner based on lowest RMSE and highest R² to predict Live Discharges. Out of the four regression learners that were chosen for final selection, Linear Support Vector Machines emerged as the best learner, and it did predict Live Discharges after the surge.