Novel alarm correlation analysis system based on association rules mining in telecommunication networks

Scalable out-of-core itemset mining

“…In addition, longer average transaction/maximal pattern lengths should result in an increasingly branched tree (as more item permutations are possible), thus boosting the occurrence of unitary support chains, which can benefit from the VNode representation. Datasets in blocks three and four are designed to assess the scalability of our approach, with respect to both dataset size (datasets [11][12][13][14][15][16] and average transaction/maximal pattern length (datasets [17][18][19][20][21][22].…”

“…The first attempt to perform itemset mining [3] was focused on discovering frequent itemsets, i.e., patterns whose observed frequency of occurrence in the source data is above a given threshold. Frequent itemsets find application in a number of real-life contexts, such as market basket data [3], recommendation systems [19], and telecommunication networks [20]. Frequent itemset mining algorithms have traditionally addressed time scalability, with increasingly efficient solutions that limit the combinatorial complexity of this problem by effectively pruning the search space.…”

Determination of Association Rules with Market Basket Analysis: Application in the Retail Sector

“…Timor and Şimşek determined factors affecting customers' buying behavior with decision trees in one of Turkey's largest chain of markets operating in the retail sector, by analyzing association rules on customer shopping records for a period of four months in 2004 [9]. Li and Li proposed a new system of alarm correlation analysis based on association mining to analyze the relationships between failures in a telecommunication company [10]. In his work on three different internet shopping markets in Taiwan, Chiang used the supervised Apriori algorithm to learn valuable customers and developed the RFMDR model [11].…”

A framework for mining interesting high utility patterns with a strong frequency affinity

“…Traditional frequent pattern mining considers an equal profit/weight for all items and only a binary occurrence (0/ 1) of the items in one transaction. Some research has been performed for weight-based pattern mining [45,34,38,46,3,24] by considering the different weights of items and their 0/1 appearance in each transaction. In the following two subsections, we describe some affinity/correlation-based research in the areas of frequent and weighted pattern mining.…”