Journal
EXPERT SYSTEMS
Volume 38, Issue 5, Pages -Publisher
WILEY
DOI: 10.1111/exsy.12551
Keywords
probability model; pruning; uncertain database; weighted probabilistic frequent itemset
Funding
- National Nature Science Foundation of China [61672379, 61370198, 61300187, 61370199]
- Liaoning Provincial Nature Science Foundation of China [2019-MS-028]
- China Postdoctoral Science Foundation [2018M640239]
Ask authors/readers for more resources
The study of frequent itemsets mining in uncertain databases poses a challenging problem as traditional methods cannot be directly applied. Researchers have proposed a novel weighted probabilistic frequent itemsets (w-PFIs) algorithm and achieved promising experimental results.
Uncertain data mining has attracted so much interest in many emerging applications over the past decade. An issue of particular interest is to discover the frequent itemsets in uncertain databases. As an item would not appear in a transaction of such database for certain, several probability models are presented to measure the frequency of an itemset, and the frequent itemset over probabilistic data generally has two different definitions: the expected support-based frequent itemset and probabilistic frequent itemset. Meanwhile, it is noted that the frequency itself cannot identify useful or meaningful patterns in some scenarios. Other measures such as the importance of items should be also taken into account. To this end, some studies recently have been done on weighted (importance) frequent itemset mining in uncertain databases. However, they are only designed for the expected support-based frequent itemset, and suffer from low efficiency due to generating too many frequent itemset candidates. To address this issue, we propose a novel weighted probabilistic frequent itemsets (w-PFIs) algorithm. Moreover, we derive a probability model for the support of a w-PFI candidate in our method and present three pruning techniques to narrow the search space and remove the unpromising candidates immediately. Extensive experiments have been conducted on both real and synthetic datasets, to evaluate the performance of our w-PFI algorithm in terms of runtime, accuracy and scalability. Results show that our algorithm yields the best performance among the existing algorithms.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available