CLTD-LP: an optimized top-down clustering approach with linear prefix trees for scalable frequent pattern discovery...

Area of Science:

• Data Mining
• Machine Learning
• Algorithm Optimization

Background:

• Frequent itemset and association rule extraction is crucial in data mining.
• Current Linear Prefix (LP) growth algorithms use a bottom-up approach requiring conditional pattern bases and LP-trees, which can be inefficient.

Purpose of the Study:

• To propose a novel Linear Prefix tree utilising a Top-Down Approach with Clustering (CLTDLP) method.
• To address the limitations of existing LP-growth algorithms, specifically regarding execution time and memory utilization.

Main Methods:

• The CLTD-LP algorithm employs a top-down methodology.
• It generates a subheader table to efficiently mine common items.
• This approach avoids the creation of a conditional pattern base or LP-tree.

Main Results:

• The CLTD-LP algorithm demonstrates enhanced efficiency in execution time and memory utilization.
• Across three benchmark datasets, CLTD-LP consistently outperformed existing algorithms like LP-growth, OFIM, and SSFIM.
• A significant average reduction in runtime and memory was observed.

Conclusions:

• The proposed CLTD-LP algorithm offers a more advantageous approach to frequent itemset mining.
• It provides substantial improvements in performance metrics compared to current methods.
• CLTD-LP is a promising advancement for efficient data mining tasks.