GSA Annual Meeting in Seattle, Washington, USA - 2017

Paper No. 99-8
Presentation Time: 9:45 AM

ANNA – A FUZZY LOGIC-BASED MACHINE LEARNING APPROACH TO FUNCTIONAL ECOLOGICAL MACROEVOLUTION


DICK, Daniel, Department of Chemical and Physical Sciences, University of Toronto Mississauga, 3359 Mississauga Road, Toronto, ON L5L 1C6, Canada and LAFLAMME, Marc, Department of Chemical and Physical Sciences, University of Toronto Mississauga, 3359 Mississauga Road, Mississauga, ON L5L 1C6, Canada, daniel.dick@mail.utoronto.ca

ANNA (Automated Niche Modelling for Numerical Analysis) is a recently developed method which uses a combination of techniques from machine learning and fuzzy logic to quantify functional ecological evolution on a macroevolutionary scale. Current approaches to functional ecology on the macroevolutionary scale (quantitative ecospace methods) are unable to explicitly account for niche overlap and redundancy, due to the way ecospace is quantified. This is in spite of the importance of these concepts for accurately testing the common hypothesis that empty ecospace drives macroevolutionary trends such as adaptive radiations. Furthermore, the differences (across environments and over geological time) observed between ecosystems are such that a single pre-existing ecospace framework risks overfitting. ANNA allows users to circumvent this by using methods from machine learning to create ecospace frameworks from patterns in the dataset, allowing for quick and efficient ecospace model development without building multiple ecospace frameworks a priori. ANNA is a machine learning program which quantifies functional ecological similarity, and uses this information to classify new taxa into cluster-based ecospace frameworks. It creates clusters using a Gower dissimilarity coefficient-based approach to the k-medoids algorithm, and uses a form of fuzzy discriminant analysis to classify new taxa into these clusters, based on minimal Gower dissimilarity with a fuzzy threshold. By using fuzzy membership functions, ANNA can accurately classify taxa which are highly dissimilar (outliers with respect to all clusters), taxa which are fully redundant (100% similarity to those in a given cluster), and taxa in-between, which represent degrees of niche overlap. This approach can be used to classify taxa from different geological periods or different environments into later/other ecosystems, as a metric approach to comparing ecological disparity. This can serve as a test of various models of the role of empty ecospace in major evolutionary trends, or be used to investigate how ecosystems respond to global perturbations such as environmentally-driven mass extinctions.