A Method for Knowledge Discovery and Development with Health Data

One of the most overlooked problems in the field of knowledge discovery is the acquisition and incorporation of existing knowledge about the data being analysed (Fayyad, Piatetsky-Shapiro et al. 1996; Pohle 2003; Kotsifakos, Marketos et al. 2008; Marinica and Guillet 2009). Doing this efficiently and effectively can greatly improve the relevance and usefulness of the results discovered, particularly for complex domains with a large amount of existing knowledge (Adejuwon & Mosavi, 2010; C. Zhang, Yu, & Bell, 2009). This study applies the successful Multiple Classification Ripple Down Rules (MCRDR) knowledge acquisition method to build a knowledge base from a complex dataset of lung function data, and describes a method for utilising the dataset to provide additional knowledge validation. The method acquired knowledge successfully, but indicated that a focus on rule-driven knowledge acquisition may adversely affect the MCRDR process. Knowledge acquisition was performed with multiple domain experts, with separate knowledge bases successfully consolidated using an evidence-based method to quantify differences and resolve conflicts. This knowledge comparison method was also tested as a learning and assessment tool for a small group of medical students, with positive results. In addition, the consolidated expert knowledge base was applied to the analysis of the lung function data, with a set of common data mining techniques, to reproduce and expand on a group of published lung function studies. Results showed that new knowledge could be discovered effectively and efficiently in a complex domain, despite the user having little domain knowledge themselves. Results were supported by recent literature, and include findings that may be of interest in the respiratory field. Notably, newly discovered knowledge is automatically incorporated into the knowledge base, allowing incremental knowledge discovery and easy application of those discoveries.