AN IMPROVISED METHOD AND TECHNIQUE FOR CLASSIFICATION USING ASSOCIATION RULE MINING

Abstract

Association rule mining and classification are two important data mining techniques in the knowledge discovery process. The integration of these two techniques is an important research focus and has many applications in data mining. The integration of these two techniques created new approaches called Class Association Rule Mining or Associative Classification Technique. The two combined approaches provide better classification accuracy when classifying data. Content-based information retrieval research areas require high efficiency and performance. In these applications, association rule mining detects association patterns from data, and we classify target classes based on the association patterns. Our paper mainly focuses on the combination of classification and association rule mining for accurate data classification. In this paper, we proposed to implement two new algorithms CPAR (Classification Based on Predictive Association Rule) and CMAR (Classification Based on Multiple-class Association Rules), which combine the advantages of both associative classification and traditional rule-based classification. Instead of producing a large number of frequent item rules as in associative classification, CPAR adopts a greedy search algorithm to produce rules directly from the training data. In addition, CPAR generates and tests more rules than traditional rule-based classifiers to avoid missing important rules. To avoid over fitting, CPAR uses the expected accuracy to evaluate each rule and uses the best k rules in prediction. CMAR applies a CR-tree structure to efficiently store and retrieve the mined association rules and efficiently prunes the rules based on confidence, correlation, and database coverage. Classification is performed based on a weighted χ2 analysis using several strong association rules. The extensive experiments show that CMAR is consistent, highly effective in classifying different kinds of databases, and has better average classification accuracy compared to FOIL (First Order Inductive Learner) and PRM (Predictive Rule Mining). The proposed algorithms are better in terms of memory requirements, time consumption and eliminate intermediate data structures during implementation.