Predictive Archaeological Modeling Using GIS-Based Fuzzy Set Estimation

trbA Case Study of Woodford County, Kentucky

Authors: Bailey, Keiron; Grossardt, Ted H; Ripy, John; Mink, Philip; Shields, Carl; Davis, Dan; and Hixon, James

Transportation Research Board Annual Meeting 2009, Paper #09-2475

Abstract

“Analytic predictive archaeological models can have great utility for state Departments of Transportation, but it is difficult to model the likelihood of prehistoric settlement using geographical proxy predictor variables because of the complexity of how settlement choices were actually made, and the complex interaction between these variables using GIS. In many cases classic statistical modeling approaches require too much data to be useful. This research reports on a preliminary predictive model that combines spatial analysis and fuzzy logic modeling to capture expert archaeological knowledge and convert this into predictive surface. A test area was defined in Woodford County, Kentucky, and five influencing factors were defined and calculated using the ArcGIS platform. Points were sampled and probabilities estimated using both small and large group structured processes from a broad range of archeologists that fed a forward-backward fuzzy logic induction process. It was used to generate and refine a knowledge base that mapped all inputs to an output probability function. These data were extracted from the fuzzy logic model to a lookup table and then geocoded into the ArcGIS platform, generating output surfaces showing the probability of encountering artifacts across the entire study area. The predictive results were tested using a blind control protocol with known archaeological data and established model testing statistics. The six models delivered predictive efficiencies that equaled and exceeded comparable statistical predictive models while using a much smaller number of variables as inputs.”