An R-based machine learning workflow with applications in personalized medicine

Abstract

We have developed a learning resource demonstrating the use of the tidymodels ecosystem, a collection of machine learning libraries for the R software language, in building a machine learning service with a focus on personalized medicine. The data used for the demonstration was obtained from the International Stroke Trial database that resulted from a stroke trial involving 19,435 patients from 467 hospitals in 36 countries. The stroke trial was established to study whether the clinical course of acute ischemic stroke is influenced by the early administration of aspirin, heparin, both, or neither. The learning resource provides sufficient background on the machine learning concepts and the interpretation of the results to aid the understanding of the overall machine learning process that involves exploratory data analysis, model building, model assessment, model interpretation, and model deployment. The code and the live versions of the output generated as part of this resource are hosted online at https://github.com/ashirwad/stat599-creative-component.