Confabulation-inspired association rule mining for rare and frequent itemsets

Area of Science:

• Data Mining
• Machine Learning
• Artificial Intelligence

Background:

• Association rule mining is crucial for discovering relationships in data.
• Existing algorithms can be inefficient, especially with infrequent items or large datasets.
• Associative classification presents challenges in handling imbalanced data.

Purpose of the Study:

• To introduce a new confabulation-inspired association rule mining (CARM) algorithm.
• To enhance efficiency and effectiveness in mining association rules, particularly for associative classification.
• To address the challenge of classifying minority classes in unbalanced datasets.

Main Methods:

• Developed a novel CARM algorithm utilizing a cogency-inspired interestingness measure.
• Implemented a single-pass approach for efficient rule mining.
• Evaluated the algorithm on synthetic and real-world benchmark datasets from the UC Irvine machine learning repository.

Main Results:

• The CARM algorithm demonstrates superior speed and reduced memory consumption compared to the Conditional Frequent Patterns growth algorithm.
• CARM effectively handles infrequent items due to its cogency-inspired approach.
• Statistical analysis confirms the algorithm's advantage in classifying minority classes within unbalanced datasets.

Conclusions:

• The proposed CARM algorithm offers a significant improvement in efficiency and performance for association rule mining and associative classification.
• CARM provides a robust solution for analyzing complex datasets, especially those with imbalanced class distributions.
• The algorithm's one-pass nature and focus on cogency make it a valuable tool for data mining practitioners.