Mining High Utility Sequences with a Novel Utility Function

Mining high utility sequential patterns (HUSP) is a popular data mining task. The goal is to find all subsequences that yield a high utility (e.g. high profit) in a quantitative sequence database (QSDB). Traditional algorithms for this task have many uses but a major limitation is that they rely on the maximum or minimum utility measures for calculating the utility of a pattern, thus assuming either a best or worst case scenario. These measures are unsuitable for many real-life applications such as business decision-making. To address this issue, this paper introduces a novel utility function (NUF) to calculate the utility of a sequence in each input sequence, which provides a trade-off between the above two extreme cases. A novel upper bound on NUF is designed as well as search space pruning strategies to eliminate unpromising candidate patterns early. These contributions are integrated into a novel efficient algorithm named FHNewUSM to discover frequent HUSPs with NUF. An experimental study with both real-life and synthetic databases shows that the proposed algorithm is efficient for mining HUSPs with NUF in terms of execution time, memory consumption and scalability.