Integration of Domain Knowledge into the Data Mining Process

Abstract

A recurring problem in healthcare systems across the world is the amount of missed appointments of outpatients. Apart from the economic impact for healthcare institutions, it also hinders the access of other patients to the health system. Therefore, missing appointments negatively affect the efficiency, accessibility, and delivery of healthcare. In practice it can be observed that some medical institutions try to tackle this issue by defining and applying predictive models for missing medical appointments to react on them. An essential drawback of successful AI models is, that the resulting models will not be able to explain their decisions and actions to humans. This is a key pain point for data scientists during the whole data mining process and a crucial aspect when trust in a model’s prediction is critical. A further challenge in the selected domain is that clinical staff often reports elevated workloads and are therefore limited in time for participating actively in such data mining projects. This master thesis provides an approach that describes how a domain knowledge base can be integrated into the data mining process for build of a predictive model for missing medical appointments. Its usage should ensure the acceptance of the outcoming models and make interactions with domain experts more efficient....