Mining Prominent Closed Frequent Item sets from Data Streams using Dynamic and Adaptive Minimum Support Threshold

Abstract

Prominent closed frequent itemsets (CFIs) are the set of highly frequent closed itemsets whose support is considerably larger than remaining closed itemsets in the given transactional dataset. The support value between the prominent CFIs and remaining closed itemsets make the optimum minimum support threshold as it segregates frequent and infrequent item-sets. However, finding an optimum minimum support threshold for mining prominent CFIs is not known prior especially when the data arriving as a stream. Also, using a static minimum support threshold for mining CFIs from the data stream may lead to missing of some itemsets due to concept drift and lossy counting. In this paper, the authors propose to use dynamic and adaptive minimum support threshold for mining prominent CFIs in the landmark streaming setting. This work is first of its kind and closely relate to Top-K CFI mining techniques. The novelty is in using Jenks natural break technique to dynamically find an optimum minimum support threshold for a landmark window and adaptation to the concept drift through delayed insertion. The proposed method has experimented on some well-known real and synthetic datasets demonstrating its suitability for mining prominent CFIs in transactional data streams.