Matrix decomposition algorithms for feature extraction

Abstract

Clinical decision support software is a delicate system which, can potentially be the physician’s closest friend. The aim of such systems is to be able to cleverly recommend a list of treatment options which closely matches the patient. We envisage a system which learns from experts without ever needing to ask them for feedback, and thus one which learns from past patient encounters. The system needs to be adaptive as well as dynamic, since all patients are different even if they may exhibit very similar symptoms. This paper proposes using matrices to capture such data, and algorithms using Singular Value Decomposition to predict treatments.