A Fuzzy Association Rule Mining Expert-Driven Approach To Knowledge Acquisition

ABSTRACT

This study tackles two concerns of knowledge engineers in designing and developing a fuzzy rule-based expert system (FES). First is to acquire a knowledge-base that emulates human perception of application domain concept in order to avoid sharp boundary problems. Second is the need for modelling a comprehensive fuzzy rulebased expert system which eliminates redundant rules in order to solve the problem of rule-base unwieldiness and provide for knowledge-base instant updates. This thesis introduces an expert-driven knowledge discovery approach- Fuzzy Association Rule Mining Expert-Driven (FARME-D) approach to knowledge acquisition. In doing this, the Apriori-like Fuzzy Association Rule Mining algorithm was adopted for mining historical databases based on expert-driven approach (where the interval boundaries, fuzzy sets membership function model and fuzzy rules consequences are determined by the expert’s opinion about the domain data). The fuzzy models were constructed using trapezoidal (trapmf) and triangular (trimf) membership functions based on the domain expert description of the database and literature. The implementation was done using C# programming language. The novelty of this approach was demonstrated by developing a prototype fuzzy expert system with mining generated rules using a case study of Coronary Heart Disease (CHD) as a cardiovascular disease in medical domain. FARME-D approach generated 79 rules as against 108 rules by standard rule-base formulation approach. Using a test case approach of validation, it was observed that FARME-D approach saved 20% of memory size utilized by the knowledge-base and achieved 27 % rule deduction while the accuracy is maintained. The statistical analysis of the result, using t-test and ANOVA revealed that decision making by FARME-D approach is significantly not different from the result by standard rule-base formulation and the domain expert at 95% confidence.

In conclusion, adopting FARME-D automated knowledge acquisition in modelling fuzzy expert system enhances the system comprehensibility by eliminating redundant rules and save memory usage. The rules generated based on expert-driven approach correspond to human perception of the application domain as compared to data-driven approach. Also, the integration of FARME-D approach to standard fuzzy expert system architecture provides for knowledge-base instant updates and resulted in a novel architecture called Fuzzy Association Rule Mining Expert System (FARMES). In future research, the mining process could be extended to involve text mining, image mining, voice mining and web mining in order to extend the scope of knowledge acquisition which will turn out to enrich the knowledge-base. Also, the knowledge representation could be extended beyond production rule to semantic net and case bases representations.