Applying Genetic Algorithm and Weight Item to Association Rule

• Main Authors: Shu-Hsien Kung, 龔書賢
• Other Authors: Chieh-Yuan Tsai
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/28152806321824234896

碩士 === 元智大學 === 工業工程與管理學系 === 91 === Association rule is one of the most important and useful technologies in data mining applications. Association rule technologies extract unknown information from large database and summarize meaningful relation among items to help business managers make better decision. Currently, most of the technologies are focused on basket analysis in supermarkets. Although the usage of the technologies does improve the understanding of item relationships, most of these works consider only buying or not-buying behavior. They did not consider important item properties such as profits to raise the quality of generated association rules. In addition, deciding suitable threshold values of support and confidence is critical to the quality of association rule technology. However, there is few researches focus on how to decide the threshold values. In this thesis, a new association rule algorithm is introduced to solve the above limitations. In this algorithm we use the concepts of temporary support and data index to represent association rules that are transformed into binary formats. To emphases other important item properties, this research uses the weighted items to represent the importance of individual items. These weighted items are used into the fitness function of heuristic genetic algorithms (GA) to estimate the value of different rules. The genetic algorithms can generate suitable threshold values for association rule mining. The method proposed in this thesis is successfully applied to several retailing transaction databases and one real-world credit card database. These applications show that weighted items apply in fitness function of a genetic algorithm can estimate the value of association rules efficiently. It is also found that genetic algorithms can really suggest suitable threshold values to get quality rules. These results demonstrate that the proposed algorithm is a practical method for increasing the quality of generated association rules.