DYNAMIC CLASSIFICATION OF STREAMING DATA USING PREDICTIVE ASSOCIATION RULES

Abstract

Classification based on predictive association rules (CPAR) is a classification method which combines the effectiveness of both association rule making and rule-based classification. For rule generation process, CPAR is more effective than traditional rule-based classification algorithms because repeated calculation is avoided and multiple literals can be selected to generate multiple rules in same time. CPAR inherits the basic ideas of FOIL (First Order Inductive Learner) algorithm and PRM (Predictive Rule Mining) algorithm in rule generation. In this dissertation entitled "DYNAMIC CLASSIFICATION OF STREAMING DATA USING PREDICTIVE ASSOCIATION RULES ", a framework is proposed to make CPAR algorithm work for dynamic streaming data. In classification part of CPAR algorithm, if any data tuple is not satisfied by any rule's antecedent then that data tuple is treated as unclassified data tuple. In proposed framework, a post processing step is added in CPAR algorithm. In post processing, unclassified data tuples are buffered in bucket of different sizes. When bucket is full then it is analyzed to find interesting patterns emerging. These interesting patterns are transformed into rules with help of domain knowledge. These newly generated rules are inserted in rule base. This modified rule base will be more enhanced to handle data tuple of new emerging class patterns. So by implementing this approach, CPAR is modified as dynamic classification using association rules (DCPAR) algorithm which can work for streaming data.