A study of fitness functions in evolutionary algorithms on symptom-herb relationship discovery

Chinese Medicine Formulae (CMF) is the most common treatment by prescribing sets of herbs to address the patient´s syndromes and symptoms in Traditional Chinese Medicine (TCM). Genetic Algorithm (GA) was used to model the discovery process in this paper. Since a fitness function plays a key role in GA, the definition of such a function carries a specific semantic meaning and may yield different results. A change in the fitness function will create a different landscape for the GA to solve. In other words, each function will generate different answers for a given dataset. This paper aims to understand the impact and the appropriateness of a fitness function to the discovery of symptom-herb relationship. Four different fitness functions were applied to an insomnia dataset. These fitness functions use either the frequency or efficacy as metrics. Some fitness functions were successful to extract core and peripheral sets from the datasets. Some of the results were even shown to be statistical significant, however, not all functions are appropriate for the discovery process. The impact and implications of crafting a fitness function should be carefully considered.