TKFIM: Top-K frequent itemset mining technique based on equivalence classes.

Area of Science:

• Data Mining
• Pattern Recognition

Background:

• Exponential data growth presents challenges for traditional frequent itemset (FI) mining algorithms.
• Existing threshold-based and size-based algorithms can miss crucial information or require complex parameter tuning.
• Support threshold selection is difficult for users unfamiliar with dataset characteristics, impacting FI mining accuracy.

Purpose of the Study:

• To propose a novel method for discovering frequent itemsets (FIs) without relying on a support threshold.
• To enhance the accuracy and efficiency of FI mining in large datasets.
• To address the computational deficiencies and potential inaccuracies of existing FI mining techniques.

Main Methods:

• Developed a Top-k frequent itemsets mining (TKFIM) approach.
• Utilized class equivalence and set-theory concepts for pattern discovery.
• Compared TKFIM against state-of-the-art methods like Top-k miner and Build Once and Mine Once (BOMO).

Main Results:

• TKFIM successfully mines all frequent itemsets without missing any patterns.
• Demonstrated significant performance gains over Top-k miner (up to 92.70%) and BOMO (up to 100%) across various datasets.
• Achieved substantial improvements in execution time and overall performance compared to existing techniques.

Conclusions:

• The proposed TKFIM method offers accurate and efficient frequent itemset mining.
• TKFIM overcomes the limitations of threshold-based approaches, particularly for users lacking dataset expertise.
• The method is suitable for large-scale datasets, providing a robust alternative for pattern discovery.