Supervised methods to support online scientific data triage

Abstract

This paper presents machine learning approaches based on supervised methods applied to triage of health and biomedical data. We discuss the applications of such approaches in three different tasks, and evaluate the usage of triage pipelines, as well as data sampling and feature selection methods to improve performance on each task. The scientific data triage systems are based on a generic and light pipeline, and yet flexible enough to perform triage on distinct data. The presented approaches were developed to be integrated as a part of web-based systems, providing real time feedback to health and biomedical professionals. All systems are publicly available as open-source.