A Comparison of Machine Learning Tools for Early Prediction of Sepsis from ICU Data

Abstract

We explore the efficacy of modern machine learning methods for the task of modeling sepsis progression. We applied a novel imputation and feature selection scheme based on signal processing technology and our medical expertise. We compared the performance of several approaches including neural networks, sparse quantile regression , and baseline classification algorithms such as random forest and SVMs. We conclude that the application of neural network, random forest, sparse quantile regression , neighborhood algorithms, and naive Bayes clas-sifiers yields superior performance with respect to accuracy , sensitivity, and specificity.