Show simple item record

dc.rights.licenseCC-BY-NC-ND
dc.contributor.advisorRenooij, S.
dc.contributor.advisorWieten, G.M.
dc.contributor.advisorBex, F.J.
dc.contributor.authorLeeuw, T.A.M.P. de
dc.date.accessioned2020-09-21T18:00:16Z
dc.date.available2020-09-21T18:00:16Z
dc.date.issued2020
dc.identifier.urihttps://studenttheses.uu.nl/handle/20.500.12932/37697
dc.description.abstractBayesian networks (BNs) are powerful mathematical tools that have found applications in many fields where uncertainty plays a role, such as medicine, forensics and law. In short, a BN consist of a graph and a set of probability parameters, and can be used to compute a probability of interest. In certain domains, data is becoming increasingly available; hence, in these domains BNs can often be constructed directly from data. In data-poor domains however, the network has to be handcrafted with the help of a domain expert. The current literature provides little guidance on manual construction of the BN graph. In this thesis we present a set of guide- lines and comparison measures that provide practical aid in building a BN graph by hand with the help of a domain expert for a problem from any domain. In contrast to earlier approaches, we aim for a set of guidelines that can be used for cases from any domain while still maintaining a high degree of practical applicability. To this end, we carry out a comprehensive literature study on current forms of guidance in this area, a detailed examination of advantageous characteristics for BN graphs, and create and assess a database of manually constructed BN graphs from the literature. To evaluate the applicability of these guidelines and measures we carry out a case study, where we take BNs constructed in the forensic and le- gal domains as our case. We nd that we largely succeeded in providing guidance for an inherently subjective process (i.e. manual BN graph construction). In addition, the BN graphs from the forensic and legal domains are generally in line with what our guidelines describe. Testing the applicability of our comparison measures however proved difficult, due to the fact that the literature provided few different BN graphs for the same case.
dc.description.sponsorshipUtrecht University
dc.format.extent3174771
dc.format.mimetypeapplication/pdf
dc.language.isoen
dc.titleFormulating guidelines for the manual construction of BN graphs in data-poor domains
dc.type.contentMaster Thesis
dc.rights.accessrightsOpen Access
dc.subject.courseuuArtificial Intelligence


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record