Stability Aware Association Rule Selection for Retail Transaction Analysis
DOI:
Keywords:
Association rules; market basket analysis; temporal stability; support; confidence; lift; pattern validation; data mining.
Abstract
Association rule mining can produce patterns with high confidence or lift that fail to remain useful when transaction conditions change. This paper develops an analytical screening framework that combines support, antecedent count, temporal stability, and a conservative confidence score. Two constructed periods of 1,000 transactions are used to compare four candidate rules. One rare rule has 90 percent confidence and a lift of 3.0 in the discovery period, but its validation confidence falls to 40 percent and its lift falls to 1.0. A more stable rule retains 68 percent confidence and a lift of 1.7 in validation. The proposed screening procedure retains that stable rule while rejecting candidates that are rare, unstable, or insufficiently informative relative to the current baseline. Wilson lower scores are used as an illustrative screening device under a binomial sampling interpretation, not as proof of statistical significance for synthetic data. The study demonstrates why pattern discovery and operational acceptance should be separate decisions and identifies the additional evidence required for real retail deployment.
Downloads
Published
Issue
Section
License

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

