Ensemble based Cost-Sensitive Feature Selection for Consolidated Knowledge Base Creation

This is paper proposes a knowledge construction system. The key objective of the system is to extract knowledge from structured data which is generally available in the form of electronic medical records (EMR). In this regard, the main focus of the research is to design and develop a domain-independent system that is capable of assisting the domain expert(s) in gaining non-trivial insights from the underlying EMR data. It is important to note that most of the research in the domain of cost-sensitive feature selection relies on black-box models which only provide a prediction of a final class label. Whereas, the goal of this research is to acquire insights for domain experts such as chronic kidney disease classification. This goal is achieved by designing and developing a knowledge construction system that is based on a two-stage methodology. Stage one deals with identifying salient cost-sensitive features in the EMR data, whereas, stage-two deals with consolidating knowledge (i.e. in the form of production rules) from a set of interpretable machine learning models. Finally, in order to demonstrate the efficacy of the system a chronic kidney disease case study is adopted.