A framework for mining of frequent patterns and class association rules from incremental data

Abstract

Data mining is the task of extracting meaningful, attractive and unseen patterns from large amounts of data. Association Rule Mining and Classification are considered as significant data analysis tasks in data mining, and they analyse the relationship between the data elements. Associative Classification (AC), which is a combination of association rule mining and classification, has emerged as an efficient classification model, and offers higher accuracy than the traditional classification methods. The algorithms of frequent pattern mining and classification assume that the databases are static, and hence, the batch-processing method is used. However, the real-time databases are usually record-based and they are incremental in nature, where the set of records is being added to the database. In this case, existing batch-processing methods do not use the previously mined information in incrementally growing databases. This motivates the need for incremental methods which maintain and update the mining results as the database grows. Particularly, in the context of frequent pattern mining and classification, maintenance of frequent patterns and class association rules is important. Hence, this research work is focused on providing effective mining methods for frequent patterns and association based classification, when the database is added incrementally. It proposes a framework for frequent pattern mining and class association rule mining from the incremental datasets. The framework consists of three phases such as frequent pattern mining from an incremental data called data stream, mining of Class Association Rules (CARs) from incremental datasets and generation of Weighted Class Association Rules (WCARs), mining of Constraint Class Association Rules (CCARs), and building of classifier by applying the rule pruning and selection techniques.In the first phase, an enhanced algorithm called Recent Frequent Pattern Mining using Diffset with Elimination of Null Transactions (RFP-DIFF-ENT) is proposed, to extract the frequent itemsets from the stream data using the sliding window model. newline