Association rule mining associates one or more attributes in a dataset to discover hidden and significant relationships between the attributes. The quality of the association rules are strongly limited by the interestingness measures and the number of the rules obtained. This paper intends to propose a technique to reduce the quantity of the rules without compromising the usefulness factor and thereby improves the computational efficiency of rule mining. The proposed framework reduces the number of rules by combining mining and post-mining techniques. Particle swarm optimization is used in the mining process to compute an optimal support and confidence parameters. The collection of strong rules is then obtained using these computed parameters. In the post-mining process, domain ontology is designed to map the database. Domain ontology helps in providing a formal, explicit specification of a shared conceptualization. Based on the user knowledge and the domain ontology, most interesting rules are discovered. A GUI based framework is also designed to assist the users in discovering the rules. Promising results were obtained when experiments were conducted with the Adult dataset of UCI machine learning repository.
展开▼
