A novel Fibonacci windows model for finding emerging patterns over online data stream

Abstract

Patterns i.e. the itemsets whose frequency increased significantly from one class to another are called emerging patterns (EP). Finding EP in a massive online data streaming is a tough yet complex task. On one hand the emergence of patterns must be examined at different time stamps since no one knows when the patterns may be emerging; on another hand, EP must be found in a given limited time and memory resources. In this work a novel method to accomplish such task is proposed. The history of itemsets and their support is kept in a novel data window model, called Fibonacci windows model, which shrinks a big number of data historical windows into a considerable much smaller number of windows. The emergence of itemsets being extracted from online transactions is examined directly with respect to the Fibonacci windows. Furthermore, as the historical windows are recorded, EP can be found both in online and offline mode.