Doctoral Degrees (Information Science)

Permanent URI for this collection

https://scholar.sun.ac.za/handle/10019.1/627

Browse

Browsing Doctoral Degrees (Information Science) by Subject "Algorithmic knowledge discovery"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Knowledge discovery and anomalies — towards a dynamic decision-making model for medical informatics
    (Stellenbosch : Stellenbosch University, 2018-03) Arndt, Heidi; Kinghorn, Johann; Stellenbosch University. Faculty of Arts and Social Sciences. Dept. of Information Science.
    ENGLISH SUMMARY : Worldwide healthcare has become a major concern for modern society, which is challenged to make quality care accessible and affordable to all. With a slowing world economy, rapidly rising medical costs and a better-informed customer base, governments and healthcare organisations are under pressure to deliver a product that focuses on quality care, transparent costs and an excellent patient experience. This requires well-informed and nimble operating and decision-making by healthcare organisations, putting pressure on the discipline of informatics within systems. In a comprehensive literature survey, it was found that healthcare organisations are organisations made up of a wide variety of subsystems operating in a complex environment. In addition, there are individualities that challenge the development of health information systems. Bisociative knowledge discovery, which is the creative discovery of previously unknown information from habitually incompatible domains, was introduced as an alternative tool to address the need for decision support in the healthcare sector. It was further found that information networks are a useful way to integrate data from habitually incompatible domains. Lastly, frequent pattern mining was identified as the machine learning tool for mining bisociations within information networks. A knowledge discovery framework for data-intensive research focusing on the field of biomedical informatics was developed in this study. Within this framework, data are represented as integrated, heterogeneous information networks, and machine learning algorithms are applied to the data with the explicit purpose of finding interconnectedness within these structures that can lead to bisociative knowledge discoveries. This framework was further developed into a knowledge discovery process model for bisociative knowledge discovery with a focus on the healthcare sector. The knowledge discovery process model for bisociative knowledge discovery was then applied in a case study which made use of the Nationwide Inpatient Sample data that forms part of the Healthcare Cost and Utilization Project. The case study successfully demonstrated the construction of habitually incompatible domains and their integration into a heterogeneous information network. Furthermore, it demonstrated the application of frequent pattern mining algorithms to extract subgraphs from the constructed information network. This was followed by the constructing of the extracted subgraphs as concept graphs with the purpose of visualising the results for further interpretation. At the end of this research it was concluded that:  The proposed explorative data mining method using bisociative knowledge discovery revealed unexpected, potentially interesting relationships within the constructed information network.  Modelling data from the healthcare sector as an information network allowed visual insights into the structure of the data, which supported the detection of novel insights that otherwise would not have been revealed.  Organisations operating in a complex environment can be successfully unpacked into rich layers of abstraction and the integration of these layers can be automated through computing.