Appropriate use of machine learning in healthcare
Bunyamin Ozaydin, Eta S. Berner, James J. Cimino
Abstract
Machine learning methods, a subdomain of artificial intelligence, in healthcare have been experiencing rapid growth and development but these methods have also been criticized. This article explores some of the more frequent criticisms leveled at machine learning approaches and offers suggestions on how to address problems of bias, “black box” systems, appropriate training for users, and monitoring model performance. Some of these problems can potentially be addressed using machine learning methods. We advocate for more involvement of clinicians in the development and evaluation of machine learning systems in healthcare and more regular monitoring of performance to take advantage of the strengths and minimize the weaknesses of these systems.