A Machine Learning Pipeline Using KNIME to Predict Hospital Admission in the MIMIC-IV Database

It is well known that overcrowding in emergency department (ED) lowers the standard of care and raises the risk of medical errors. An initial predictive supplementary tool of hospital admission at an early stage of a patient's arrival to the emergency department (ED) can provide health care professionals a number of advantages, such as, more efficient patient flow management and better hospital care. In this paper, we use data from the Medical Information Mart for Intensive Care IV Emergency Department (MIMIC-IV-ED) database to predict whether a patient will be admitted to the hospital or not. The choice of predictive attributes was driven by simplicity (a set of basic vital signs were used) so that the prediction can be made at an early stage of the patient's arrival. Several versions of Machine Learning (ML) algorithms based on Decision Trees (DT) were used for classification and prediction. An important asset of the proposed methodology is that the whole process is implemented through an ML pipeline created with an open-source, visual programming tool. The proposed methodology contains the pre-processing stage, the modelling stage includes seven classifiers, and the combined visualization of the evaluation of the predictive models. The Gradient Boosted Trees method outperforms the rest of the algorithms that were used. An accuracy of 80% can be achieved only by using early triage data.