Uncovering Complex Stroke Risk Patterns Using Association Rule Mining on Synthetic and Real-World Health Data

Stroke remains a major global cause of morbidity and mortality, with a necessity for methods to uncover complex relationships among risk factors. This study uses Association Rule Mining (ARM) for the detection of co-occurring patterns of stroke risk through Apriori and FP-Growth algorithms on large-scale synthetic ($\mathbf{n} \boldsymbol{=} \mathbf{5 0, 0 0 0}$) and small-scale real-world $(\mathrm{n}=5,110)$ health datasets. Our objectives were to extract interpretable risk patterns, contrast algorithm performance, and ascertain the utility of synthetic data for exploratory analysis. Interesting findings from the real data were strong associations (Lift > 5, Confidence > 0.7) between factors like younger age, unknown smoking status, and unmarried status with non-stroke outcomes, potentially representing sociod-emographic factors. The synthetic data validated the less robust but clinically plausible correlations (Lift $\approx 1.02$) between obesity and high glucose levels, demonstrating its usefulness for initial pattern discovery while highlighting the need for real-world validation. FP-Growth was more computationally efficient throughout. This research has shown ARM's potential to mine health data for interpretable knowledge, informing scalable prevention programs and translating data-driven insights into actionable clinical wisdom.