Performance comparison of inertia weight and acceleration coefficients of BPSO in the context of high-utility itemset mining

One of the best performing methods in high-utility itemsets mining is based on binary particle swarm optimization. This paper aims to improve the quality of that method by finding ways to increase both the number of obtained itemsets and the associated total utility of those itemsets. The method used is based on binary particle swarm optimization, which is one of the best algorithms to obtain high-utility itemsets based on swarm intelligence. To get the maximum results we tuned up the method to determine the optimal configurations of four parameters: initial population, inertia weight, acceleration coefficients, and velocity clamping. Our experimental results on five datasets indicate that the best configuration is combining constriction factor or adaptive method for setting inertia weight and using self-adaptive for acceleration coefficient. The use of frequent itemset or transactions of the highest total utility for initializing population does not have a significant addition compared to the inertia weight and acceleration coefficient parameters. The correct configuration of the binary particle swarm optimization can increase the number of high-utility itemsets compared to the standard configuration. The percentage increase of the number of itemset compared to the existing standard method is 157.93% for chess dataset, 134.39% for connect dataset, 42.69% for accident dataset, 4.62% for foodmart dataset, and 1.58% for mushroom dataset.