Development of An Interactive Dashboard for Outcome Prediction of a Patient Length of Stay
Keywords:Machine Learning, Xtreme Gradient Boosting Regressor, Extra Trees Regressor, Random Forest Regressor, Gradient Boosting Regressor
Machine learning (ML) models predicting operative outcomes of mortality, and in-hospital morbidity e.g., patient length of stay and hospitalization costs can play an important role in examining the efficiency of healthcare services and resource allocation planning in a hospital. However, understanding the inner workings of the ML model in predicting the outcomes is challenging. Therefore, it is beneficial to build an interactive dashboard for interpretations of the ML model. In this study, a few regression-type supervised ML algorithms of Xtreme Gradient Boosting Regressor, Extra Trees Regressor, Random Forest Regressor, Gradient Boosting Regressor and Neural Network were analyzed to predict the patient length of stay (LOS) by using the MIMIC3d dataset. Then the finding of prediction performance was evaluated in terms of root mean squared error (RMSE) and the R-squared (R2) performance indexes. The performance results of the ML model evaluation were compared, and the best ML model which was Xtreme Gradient Boosting Regressor with RMSE of 3.44 and R2 of 0.992 was used to create an interactive dashboard of outcome prediction for a patient length of stay. All these works were done using Python programming. This dashboard was successfully tested in a local host, and it could be used by hospital administrators to conduct ML analysis and prediction of patient LOS patients helping them to optimize resource allocation planning in hospital units.