Discovering highly expected utility itemsets for revenue prediction

Identifying patterns of items that are purchased frequently and generate high profits is crucial for inventory and profit management. However, neither approaches based on frequent itemsets nor those based on high-utility itemsets (HUIs) can meet this requirement alone. Therefore, we propose a new approach, named the FIHUM algorithm, for identifying frequent HUIs. The novel characteristic of the FIHUM algorithm is that it can effectively identify frequent itemsets with high utility (frequent HUIs) without generating many high-utility candidate itemsets. Moreover, experimental results from retail data sets reveal that the FIHUM algorithm integrates the advantages of frequent itemsets and HUIs. Finally, the highly expected utility itemsets (frequent HUIs) generated using the FIHUM algorithm are suitable for predicting patterns of items that are purchased frequently by customers and generate high profits in next-period transactions.