Geo-spatial modelling with unbalanced data:modelling the spatial pattern of human activityduring the Stone Age

Abstract

With the increasing availability of data, geoscienceprovides many methods to model the spatial extentof various phenomena.Acquiring representative, highquality data is the most important criterion to assess thevalue of any spatial analysis, however, there are many situationsin which these criteria cannot be fulfilled. Archiveddata, collected in the past, for which analysis cannot berepeated or supplemented is a very common informationsource. Archaeological data collected at a regional extentduring years of field work and superficial observations arean additional example. Such data rarely provide representativesamples and are usually imbalanced; only very fewexamples contain useful data, while many examples remainwithout any archaeological traces. In spite of theselimitations archaeological information presented in theform of maps can be a useful and helpful tool to analysethe spatial patterns of some phenomena and, from a morepractical point of view, a tool to predict the location ofundiscovered occurrences. The primary goal of this paperis to present a methodology for modelling spatial patternsbased on imbalanced categorical data which do not fulfilthe criteria of spatial representation and incorporatesuncertainty in its decision process. This concept will bediscussed using a collection of Stone Age sites and set ofenvironmental variables from the postglacial lowlands inWestern Poland. We will propose a machine-learning systemwhich adopts CART through bootstrap simulation toincorporate uncertainty into the spatial model and utilisethat uncertainty in the decision-making process. Finally,we will describe the relationships between the model andenvironmental variables and present our results in cartographicform using the principles of decision-tree cartography.

Keywords

• uncertainty
• spatial modelling
• machine learning
• stone age
• geomorphometry