Clinical data analysis using ontology-guided rule learning

Currently, research in Machine Learning (ML) mainly focuses on the ability to process very large amounts of data and build accurate models. Problems related to complexity, heterogeneity, and semantics of healthcare data are often out of the main focus. Healthcare is particularly rich in background knowledge. Surprisingly, few ML methods used in healthcare can handle these sources of background knowledge, and instead treat healthcare data as a set of numbers without particular meaning. This paper explores an approach that can fill in this gap. A medical ontology (i.e., UMLS) is proposed to provide background knowledge for the ML method to understand healthcare data. The ontology-guided ML-based rule induction method is described and illustrated to analyze the clinical data supplemented with an ontology-based background knowledge.