A Hybrid PSO with Repetitive Local Search to Optimize Frequent Pattern Mining

Abstract

Frequent pattern mining is the core operation for expanding the scope of data analysis. The
dataset consists of sub-sequences, itemsets, and sub-structures. These items that occur
frequently more than the user-specified value (support) is considered for the mining process.
Abundant research has been carried out in the frequent itemset mining due to the involvement
of huge database instances. Dealing with a large transactional database is a challenging
problem in FPM. When handling a huge quantity of data, another challenging issue is to find
more number of frequent itemsets efficiently. Evolutionary computing is an interesting
technique applied to solve the frequent itemset mining problem by researchers. The issues
encountered in the existing PSO strategies are premature convergence; limited exploration of
the dimensional space provides low-quality solutions. Hence, an improved hybrid version of
PSO is applied to optimize the FPM process. First, the frequent itemset mining problem is
modeled from the perspective of the PSO algorithm. Second, in the PSO algorithm, the
convergence rate is maximized using the constriction factor approach. Third, to obtain good
quality solutions, hybrid PSO with repetitive local search is utilized to maintain the balance
between intensification and diversification. Finally, the proposed approaches are analyzed on
the basis of standard solution, convergence rate, and computational time. According to
experimental findings, the hybrid PSO technique significantly enhanced the frequent itemset's
quality and convergence rate. Overall, the hybrid PSO algorithm performs better than the
existing algorithms.