CONTEXT-AWARE ASSOCIATION RULE MINING FOR EFFECTIVE KNOWLEDGE EXTRACTION IN CARDIOVASCULAR CLINICAL DECISION SUPPORT

Abstract

Coronary artery disease (CAD) is a disorder that gradually develops due to the accumulation of fatty deposits within the arteries, which limits the blood supply to the heart. It is often not detected until serious complications occur, such as blockage or heart attacks. This paper proposes a framework for extracting medical association rules from patient data (facts file) using three advanced techniques—FP-Growth, Eclat, and HUIM—after cleaning, filtering, and converting the JSON fact file into a structured binary representation. FP-Growth provides efficient mining of frequent itemsets through the FP tree, while Eclat benefits from the intersections of transaction ID lists (TID-lists) within a depth-first search. Meanwhile, HUIM prioritizes "clinical utility" by focusing on rare high-impact features (such as significant biomarkers). The framework takes into account the contextual variability associated with regional factors and lifestyle in Iraq (for example, Basra vs. Baghdad). The resulting rules are filtered according to criteria: support, confidence, and lift to retain the strongest rules supporting the diagnosis. Eclat excels more than FP-Growth and HUIM; the average confidence is ~1.0 with near-zero variance, passing 100% of all confidence thresholds (0.5→1.0), with an average support of ~0.39 and a moderate lift (~1.3). The impact of the differences from FP-Growth is significant (Cohen’s d≈1.6), and the ANOVA/Tukey results confirm its statistical superiority. These characteristics make it suitable as a verification/validation layer for reliable rules.