Efficiently Discover Multi-level Maximal High-Utility Patterns from Hierarchical Databases

Patterns discovered through mining approaches often result in a large number of findings. Analyzing these patterns can be a time-consuming process. Since the early days of pattern mining, compact representations of patterns have been utilized alongside regular patterns. There are two types of compact representations: closed and maximal. Pattern mining approaches, such as frequent pattern mining (FPM) or high-utility pattern mining (HUPM), have incorporated these compact representations to reduce the number of outcomes while still preserving insights from the databases. Recently, the HUPM task has been extended to handle hierarchical transaction databases, significantly enlarging the search space compared to traditional transaction databases. Consequently, the number of discovered patterns has also increased. However, approaches have yet to be proposed to address this issue. Therefore, this study proposes new techniques for mining multi-level maximal high-utility patterns from hierarchical transaction databases. Additionally, we introduce a novel algorithm called MaxMinerML, which efficiently solves this task by leveraging these techniques. Empirical evaluations on real-world databases have demonstrated that MaxMinerML is efficient in memory usage and mining time.