Redundant or Interesting? Inspection of Redundant Association Rules in a Psychiatric Dataset
Bildiriler > Tam Metin Bildiri
Yazarlar: İnanç KABASAKAL, Yilmaz Kafali, Helin
Zenodo (CERN European Organization for Nuclear Research), , , Uluslararası, Tam Metin Bildiri
Özet
Association Rule Discovery is a process that extracts potentially useful patterns in the form of association rules. The evaluation of the rules in descriptive studies is often conducted by referring to interestingness measures, and often with the participation of domain experts, which might require significant time and effort. Reduction techniques help this challenge by filtering out redundant rules. This paper aims to analyze a list of rules obtained through a study in child psychiatry and present the redundancies discovered by reduction techniques. The case being examined involves demographic variables, symptoms, and diagnoses of psychiatric diseases regarding 263 patients from Turkey. With thresholds of 2%-60% interval for support and 10% confidence, a list of 54 rules was extracted through rule mining. In our study, the redundancy of rules that consist of three items in the form of {X=>YZ} was determined using the rule-of-thumb by [1] and the ratings assigned by a domain expert regarding the usefulness of rules were examined using statistical methods. The results demonstrated that the usefulness of original rules and their variants in the form of {X=>Y, X=>Z} were statistically different, and no such difference was found with the variants of the form {XY=>Z, XZ=>Y}.