An incremental high-utility mining algorithm with transaction insertion.

Area of Science:

• Data Mining
• Database Systems
• Machine Learning

Background:

• Traditional association-rule mining focuses solely on item frequency, limiting its applicability in real-world scenarios with varying item profits and quantities.
• High-utility mining addresses these limitations by incorporating profit and quantity measures, but existing algorithms often struggle with dynamic transaction insertions.

Purpose of the Study:

• To develop an efficient incremental algorithm for dynamic high-utility mining that effectively handles transaction insertions.
• To reduce computational overhead, specifically database rescans and the combination explosion inherent in pattern-growth mechanisms.

Main Methods:

• An efficient incremental algorithm utilizing utility-list structures to manage dynamic transaction insertions.
• Incorporation of an enumeration tree and analysis of 2-itemset relationships to accelerate computations.
• Avoidance of candidate generation, a common bottleneck in traditional high-utility mining algorithms.

Main Results:

• The proposed algorithm demonstrates reduced computational complexity for dynamic high-utility mining with transaction insertions.
• Experimental results validate the algorithm's efficiency in terms of runtime and memory consumption.
• The algorithm effectively generates high-utility patterns without requiring a full database rescan.

Conclusions:

• The developed incremental algorithm offers an efficient solution for dynamic high-utility mining, particularly in scenarios with frequent transaction updates.
• This approach significantly improves performance compared to existing methods by optimizing computational processes and avoiding redundant calculations.
• The findings contribute to more scalable and practical high-utility pattern discovery in large, evolving datasets.