Optimised Association Rule Mining for Health Data

Abstract

Association Rule Mining (ARM) has been recognised as a valuable and easy-to-interpret data mining technique in response to the exponential growth of big data. However, research on ARM techniques has mainly focused on enhancing computational efficiency while neglecting the automatic determination of threshold values for measuring the "interestingness" of items. Selecting appropriate threshold values (such as support, confidence, etc.) significantly affects the quality of the association rule mining outcomes. This study proposes an algorithm that utilises Particle Swarm Optimization (PSO) and ARM techniques to determine optimised threshold values in the health domain automatically. The algorithm was evaluated using the UCI machine learning medical database for heart disease. Results show that the proposed algorithm is capable of generating frequent itemsets and rules in an efficient manner and can detect the optimum threshold values. This research has practical implications for the health domains, as it can extract valuable results.