IIIT Hyderabad Publications |
|||||||||
|
Efficient Mining of Frequent and Periodic High-Utility Patterns From Transactional DatabasesAuthor: Yashwanth Tatikonda Date: 2021-04-27 Report no: IIIT/TH/2021/46 Advisor:P Krishna Reddy AbstractHigh-utility itemset mining (HUIM) has emerged as an important research topic since the quantity and profit factors employed to mine the high-utility itemsets (HUIs). In HUIM, utility measure used to extract high-utility itemsets from the given transactional database. An itemset called a high-utility itemset if it satisfies a user-specified minimum utility (minU til) constraint. Another distinguishing characteristic of HUIM is identifying interestingness about the periodic behavior of high-utility itemsets in the given transactional database. Finding such itemsets have many real-world applications. As an example, such knowledge can identify recurring customer purchase behavior in retail databases. The popular adoption and successful industrial application of the HUIM model suffer from the issue of computational expensiveness. Furthermore, the existing approach to extract periodic behavior of highutility itemsets fails to discover high-utility itemsets that have exhibited partial periodic behavior in the database. In this thesis, we have proposed two improved HUIM approaches. First, to improve the performance of HUIM, we have proposed a generic high-utility frequent itemset model to find all itemsets in the database that satisfy user-specified minimum support and minimum utility constraints. Two new pruning measures, named cutoff utility and suffix utility, are proposed to reduce the computational cost of finding the desired itemsets. A single phase fast algorithm, called High-Utility Frequent Itemset Miner (HUFIMi), is introduced to efficiently discover the itemsets. Experimental results demonstrate that the proposed algorithm is efficient. Second, to enable the extraction of knowledge about the partial periodic behavior of customers, we have proposed a flexible model called Partial Periodic High-Utility Itemset Mining (PPHUIM). The proposed model discovers only those interesting high-utility itemsets that occur at regular intervals in a given temporal transactional database. The proposed model allows the variation of external utilities and exploits the notion of periodic support. An efficient depth-first search algorithm, called PPHUI-Miner, has been proposed to enumerate all partial periodic high-utility itemsets in a given temporal transactional database. We have demonstrated the efficiency of the proposed approach through extensive experimental results. Overall, we have proposed improved approaches to mine high-utility frequent itemsets and partial periodic high-utility itemsets from transactional databases and demonstrated the efficiency through extensive experimental results. Full thesis: pdf Centre for Data Engineering |
||||||||
Copyright © 2009 - IIIT Hyderabad. All Rights Reserved. |