Machine Learning Methods for Mortality Prediction of Polytraumatized Patients in Intensive Care Units -Dealing with Imbalanced and High-Dimensional Data

摘要

The aim of this study is the prediction of death of polytraumatized patients based on epidemiological, clinical and health treatment variables by means of data-mining methods. The main problems to be addressed were high dimensionality and imbalanced data. Since the techniques usually used to deal with these drawbacks, as feature selection methods and sampling strategies respectively, did not provided satisfactory results, the aim of the study was to find out the data mining algorithms showing the best behavior in this kind of scenarios. The study was carried out with data from 497 patients diagnosed with severe trauma who were hospitalized in the Intensive Care Unit (ICU) of the University Hospital of Salamanca. The results of the study reveal the better behavior of multiclassifiers as compared with simple classifiers in contexts of high dimensionality and imbalanced datasets, without the need to resort to un-dersampling and oversampling strategies, which can lead to the loss of valuable data and overfitting problems respectively.