LCIM: Mining Low Cost High Utility Itemsets.

In data science, a key task is high utility itemset mining (HUIM), that is determining the values that co-occur in data and have a high utility (importance). That task is applied for instance to identify the most profitable sets of products in transactions. A shortcoming of current algorithms is that they focus on the utility of patterns, but ignore their cost (e.g. time, effort, money or other resources that are consumed). Hence, this paper defines the problem of low cost high utility itemset mining. The aim is to find patterns that have a high average utility and a low average-cost. An example application is to find patterns indicating learners’ studying patterns in an e-learning platform that result in obtaining high grades (utility) for a relatively small effort (cost). An efficient algorithm named LCIM (Low Cost Itemset Miner) is proposed to solve this problem. To reduce the search space, LCIM uses a novel lower bound on the average cost. Observations from experiments confirm that LCIM find interesting patterns and is efficient.