It is crucial for businesses to identify any revenue bottlenecks generated from itemsets (also called products). For improving revenue, business managers must determine itemsets that should be retained in or eliminated from shops. Therefore, identifying itemsets with higher values (utility andcost-benefit) from frequent itemsets with low utility is appealing. However, past studies on utility itemset mining have focused on high-utility itemsets rather than low-utility itemsets. The current study thus proposes a new algorithm to discover frequent itemsets with low utility. Experimental results obtained using two real-life datasets show that the top k itemsets determined by the proposed algorithm have higher prediction performance in the measures of utility and cost-benefit compared with other frequent itemsets determined by a traditional algorithm.
