A Matrix Factorization Approach to Health Record Data Mining

Abstract

The increasing use of standardized electronic patient records in the health- care industry over the past few years has given rise to a new field of big data analysis with goals of identifying disease correlations, subgrouping similar pa- tients, and performing medical outcome prediction. Developments in these ar- eas have huge potential to cut spending ine ciencies and boost clinical decision support. This thesis proposes a non-negative matrix factorization approach to clinical data mining, drawing analogies to studies done in the fields of text min- ing and predictive recommender systems. We review effective modifications to the standard algorithm and run experiments on a set of patient claims data.