Summary: | 碩士 === 國立高雄應用科技大學 === 電子工程系碩士班 === 104 === Association rule is an important branch of data mining. Basically, it is the study of finding association rules from given database. These rules will then be used for prediction applications. Apriori algorithm and genetic algorithm are two promising approaches for the finding of association rules. However, each of them has its strength and limitation. Aimed at improved performance and solution quality, this study devotes to a hybrid approach with a novel operator for local optimizing.
For specified support and confidence, Apriori algorithm iteratively scans given data set, and eventually leads to a complete association rule set. It could be a time demanding task if the size of the database is large. With genetic algorithms, association rules are encoded as chromosomes, which are designed to evolve toward optimal solution via repeated selection, crossover, and mutation operations. Genetic algorithm is expected to be faster, however, with a cost of incompleteness.
To address the above-mentioned dilemma, a hybrid approach with local refinement operator is proposed in this study. The proposed approach has a better compromise between computational time and solution quality. Improvement, up to 45% in time, can be achieved, as indicated in the experiment results.
|