Data mining is essentially applied to discover new knowledge from a database through an iterative process. The mining process may be time consuming for massive datasets. A widely used method related to knowledge discovery domain refers to association rule mining (ARM) approach, despite its shortcomings in mining large databases. As such, several approaches have been prescribed to unravel knowledge. Most of the proposed algorithms addressed data incremental issues, especially when a hefty amount of data are added to the database after the latest mining process. Three basic manipulation operations performed in a database include add, delete, and update. Any method devised in light of data incremental issues is bound to embed these three operations. The changing threshold is a long-standing problem within the data mining field. Since decision making refers to an active process, the threshold is indeed changeable. Accordingly, the present study proposes an algorithm that resolves the issue of rescanning a database that had been mined previously and allows retrieval of knowledge that satisfies several thresholds without the need to learn the process from scratch. The proposed approach displayed high accuracy in experimentation, as well as reduction in processing time by almost two-thirds of the original mining execution time.

Designing an efficient association rule mining (ARM) algorithm for multilevel knowledge-based transactional databases that is appropriate for real-world deployments is of paramount concern. However, dynamic decision making that needs to modify the threshold either to minimize or maximize the output knowledge certainly necessitates the extant state-of-the-art algorithms to rescan the entire database. Subsequently, the process incurs heavy computation cost and is not feasible for real-time applications. The paper addresses efficiently the problem of threshold dynamic updation for a given purpose. The paper contributes by presenting a novel ARM approach that creates an intermediate itemset and applies a threshold to extract categorical frequent itemsets with diverse threshold values. Thus, improving the overall efficiency as we no longer needs to scan the whole database. After the entire itemset is built, we are able to obtain real support without the need of rebuilding the itemset (e.g. Itemset list is intersected to obtain the actual support). Moreover, the algorithm supports to extract many frequent itemsets according to a pre-determined minimum support with an independent purpose. Additionally, the experimental results of our proposed approach demonstrate the capability to be deployed in any mining system in a fully parallel mode; consequently, increasing the efficiency of the real-time association rules discovery process. The proposed approach outperforms the extant state-of-the-art and shows promising results that reduce computation cost, increase accuracy, and produce all possible itemsets.

No abstract

No abstract