Re-induction based mining for high utility item-sets

The High Utility Itemset mining (HUIM) is an important research area in the field of data mining and knowledge discovery. HUIM aims to discover the high utility patterns from a given database, based on a utility threshold value, where the utility is a user-defined objective function. The existing HUIM algorithms fail to consider the actual behaviour of the occurrence of patterns in database. They consider all the patterns having the same utility value to be of equal importance. However, this may not always be the case, since some patterns may occur in localized clusters in the database while others can have a more uniform sequence of occurrence. The Frequent Itemset Mining (FIM) approaches also fail to address this problem since they are based on a support framework that considers only the frequency of occurrence of an itemset in the database. To address this research gap, this study introduces a novel concept of maintaining a count value of the itemsets, called re-induction count, in order to keep track of the relative occurrence of items in the database. A novel algorithm, named Ri-Miner, is proposed to mine itemsets based on both a minimum utility threshold and their re-induction count. The experimental results show that Ri-Miner outperforms existing methods by achieving a 15% improvement in execution time and a 10% reduction in memory usage. The proposed method can be useful in various applications that require capturing the underlying occurrence behaviour of the patterns the database, like market-basket analysis, healthcare, web stream analytics, etc.