A feature ranking algorithm for clustering medical data

Abstract

AbstractClustering methods are often applied to electronic medical records (EMR) data for various objectives, including the discovery of previously unrecognized disease subtypes. The abundance and redundancy of information in EMR data raises the need to identify and rank the features that are most relevant for clustering. Here we propose FRIGATE, an ensemble feature ranking algorithm for clustering, which uses game-theoretic concepts. FRIGATE derives the importance of features from solving multiple clustering problems on subgroups of features. In every such problem, a Shapley-like framework is utilized to rank a selected set of features, and multiplicative weights are employed to reduce the randomness in their selection. It outperforms extant ensemble ranking algorithms, both in solution quality and in speed. FRIGATE can improve disease understanding by enabling better subtype discovery from EMR data.