Improving rule-based classification systems by using arguments

Abstract

Improving the predictive accuracy of rule-based classifiers, specifically the Clark Niblett 2 algorithm, can be done by using an argumentation-based approach. Previous research on this topic focuses on using expert feedback to improve the predictive accuracy. This research has concentrated on providing a working algorithm and limits itself to a few domains which require expert feedback. In combination with writing this thesis an application has been built which makes the practical use of argumentation-based classification a possibility. Experi- ments are included to show the validity of the techniques in the application and to research the viability of the use of argumentation-based classifiers in a num- ber of other domains. The results show that the argumentation-based approach is solid, but possible future research into using ordered rule-based classifiers as opposed to unordered ones can provide greater benefits. The application and experiments strengthen the validity of argumentation-based classification and enhance its practical usage.