The 87th Annual Meeting of the American Association of Physical Anthropologists (2018)


Fuzzy logic as an approach for assessing population relatedness and phenotypic variation

DONOVAN M. ADAMS, REBECCA L. GEORGE and MARIN A. PILLOUD.

Department of Anthropology, University of Nevada, Reno

April 14, 2018 , Foothills Ballroom II Add to calendar

Human variation undeniably occurs along clinal gradations, with frequencies of traits and variations in size along a continuous spectrum in the human species. As such, bioarchaeological studies of biological distance have received critiques in the literature of being “typological,” while others have argued that the analysis of variation and patterns assists in understanding human relationships. To address these issues, the present research uses a novel statistical approach for biodistance analyses: fuzzy logic.

To assess the efficacy of fuzzy logic on closely related populations, the present analysis employs odontometric and dental morphological data (n=6038) collected by Tsunehiko Hanihara from fifteen populations across the globe. These data were separated by sex as preliminary analyses showed poor performance with z-scores. Raw data and principal component variables were compared and modeled in an Adaptive Neuro-Fuzzy Inference System (ANFIS) to evaluate group membership. This approach utilizes an artificial neural network to construct a system of IF-THEN rules to produce a crisp output. Fuzzy c-means analyses were also conducted to evaluate similarities between individuals and populations. Each individual is given a fuzzy output membership for each cluster, which assesses the similarity of the individual to each available cluster. ANFIS models performed poorly, with greater success at group allocation for males. C-means analyses found moderate success in identifying known population relationships and histories, particularly with females. Currently, the potential of fuzzy logic is promising; however, the method is inadequate for use on dental morphology due to issues with missing data.