Objective Interestingness Measure And Its Application In Associative Classification

With the rapidly development of information industrialization, and with quickly deepening of the application database, data mining has recently become the hotspot, in which association rule mining and classification rule mining are two of the active branches with board application. Because association rules have the features of cause and effect, which means that when the consequence of association rules represents the class label, association rules contain classification rules' features, association rule mining combining with classification rule mining can generate a novel classification method, i.e., Associative Classification.However, the traditional associative classification methods are based on frequent itemsets. They suffer from one major deficiency, that is, they often obtain a huge set of classification rules from the training data set, many of which are redundant. Also, it is challenging to build and use the classifier. Although some improved algorithms have been proposed, they still have some disadvantages.In this thesis, after current associative classification mining algorithms and interestingness measures are systemically analyzed and researched, a novel objective interestingness measure and one improved algorithms are proposed. The main tasks in the thesis are as follows:(1) The basic theories for associative classification rule mining are discussed systematically in this thesis, including the basic concepts and mining process of association rule mining and classification rule mining, and two classical associative classification mining methods. Then, the future researches of association rule mining are pointed out.(2) How to select the interested and valuable rules from a large number of association rules is one of the important contents in study of mining algorithm. Because there is a limitation in model based on support and confidence measure, the author of this thesis analyzes and researches some current interestingness measures, and puts forward a novel objective interestingness measure which is used to prune the non-interest rules in order to discover the real interested rules. The instance and experimental tests show that the novel interestingness measure is effective and practical.(3) The author of this thesis addresses some disadvantages of the popular associative classification methods and proposes a novel associative classification algorithm based on D-Miner. The novel algorithm scans database only once to mine frequent closed itemsets which would help reduce the number of classification rules. Also the novel algorithm applies the interestingness measure to prune rules, which takes much less storage space. The theoretical analysis and experimental tests show that the novel algorithm is superior in the performance.