Effective techniques for association rule mining and associative classification

Association rule mining is one of the major data mining techniques and perhaps the most common form of local-pattern discovery in unsupervised learning systems. Conventionally, an association rule is considered based mainly on the support and confidence measures which represent the usefulness and certainty of the rule respectively. However, the number of rules mined under the support-confidence framework is often too large and many of them are uninteresting. And the association rule mining process is computationally expensive as it requires exhaustive search in the huge space of possible rules. Apart from pattern discovery, association rules can also be used for data classification. Associative classification takes advantage of association rule mining in discovering a set of rules that can accurately generalize a dataset. Associative classification can achieve higher accuracy in comparison with other classification approaches. However, similar to association rule mining, the rule discovery process in associative classification is often complex and computationally expensive. And the set of association rules mined using the association rule mining framework may not be effective for direct employment by a classification task. This research aims to investigate effective techniques for enhancing association rule mining in both rule representation and mining algorithm, and discovering association rules for associative classification. As a result, we have developed a number of new techniques for association rule mining and associative classification. The contributions of this research are listed as follows: • A new type of constraint for association rules called category-based constraint is proposed. In addition, an Apriori-based algorithm called Category-based Apriori algorithm (AprioriCB) is also proposed for mining association rules with the category-based constraint. The proposed category-based constraint and AprioriCB algorithm can be used for mining interesting rules from datasets using category patterns. The AprioriCB algorithm is efficient as it can reduce considerably the search space of rules when mining interesting rules. i ATTENTION: The Singapore Copyright Act applies to the use of this document. Nanyang Technological University Library