Fast Frequent Patterns Mining by Multiple Sampling With Tight Guarantee Under Bayesian Statistics.

Area of Science:

• Data Mining
• Machine Learning
• Database Systems

Background:

• Frequent pattern (FP) mining is crucial for analyzing large datasets.
• Existing methods often overestimate sample sizes to guarantee FP quality.
• There's a need for more efficient and accurate FP mining techniques.

Purpose of the Study:

• To propose a novel algorithm, MSFP, for efficient and accurate frequent pattern mining.
• To reduce sample size requirements while maintaining theoretical guarantees.
• To improve the quality and reliability of frequent patterns identified from samples.

Main Methods:

• Developed the multiple sampling-based FPs mining (MSFP) algorithm.
• Introduced approximate frequent items (AFI) and approximate FPs without supports (AFP*).
• Utilized Bayesian statistics to stabilize pattern supports and progressive sampling for error bound improvement.

Main Results:

• MSFP effectively identifies frequent patterns using smaller sample sizes.
• The algorithm demonstrates reliability and efficiency in experimental evaluations.
• Bayesian statistics and progressive sampling enhance the accuracy of support estimation.

Conclusions:

• MSFP offers a significant improvement over traditional FP mining methods.
• The algorithm provides a theoretically sound and practically efficient approach to frequent pattern discovery.
• MSFP is a valuable tool for analyzing large datasets where sampling is necessary.