Abstract
<title>Abstract</title> <p>In line with mining association rule, frequent pattern mining is an essential process. It discovers frequently used patterns in the transaction dataset and extracts all associated rules. Numerous duplicate patterns are found when mining frequently occurring itemsets, leading to an increase in meaningless itemsets. Maximum frequent pattern and closed frequent pattern are the two techniques used in protracted frequent pattern mining. The maximal frequent pattern, regrettably, suggests that every one of its subsets is frequent and that the minimal support criterion is not crossed by their support. The disadvantage of maximal frequent pattern mining is that the precise value of the support is unknown. In closed frequent pattern mining single search is applied for the entire transaction database resulting in higher search space, consumption of memory, time and computational overhead specifically for the dense datasets. To address this limitation, a novel technique of Cluster-Aware Top-Down Linear Prefix Tree for Closed Frequent Pattern Mining (CATD-LPT-CFPM ) framework is proposed. Similar transactions from the ordered transaction database are grouped into clusters to divide the search space and to reduce the redundant processing. For each cluster separate linear prefix tree is constructed for more focused pattern research. The proposed framework adopts a multi-level pruning technique that includes cluster elimination pruning, top-down support pruning, and a novel Top-Down Closedness Pruning (TDCP) mechanism. TDCP recognize the non-closed patterns during traversal by comparing the parent and the child node support values thereby minimizing the storing and analysis of duplicate itemsets. Explored local closed patterns from each cluster are merged and verified to produce the final global closed pattern. Experimental evaluation on benchmark dataset shows that the proposed CATD-LPT-CFPM technique minimizes the number of examined nodes and organization of search space is enhanced by clustering and top-down tree traversal. While comparing the proposed CATD-LPT-CFPM with the traditional closed frequent patterns the proposed shows efficient memory storage and execution time is relatively higher due to additional clustering and prefix tree construction overhead based in datasets characteristics.</p>