Using evolutionary computation and data mining to model the emergence of archaic urban centers.

摘要

We often take our urban systems for granted. In this thesis we investigated the formation processes associated with early archaic urban centers using data mining techniques. Specifically, we investigated the emergence of Monte Alban in the Valley of Oaxaca, Mexico. One of the questions of interest is how this early city relates to models of cities based upon modern examples? For example, Doxiades suggested that an urban system had four components: the residential area, central plaza, marketplace, and transportation or circulatory network. In Monte Alban only three of the four components exist. Given the hilltop location, a marketplace is not feasible there. That leaves the three remaining components.;We extended the work of Franzel [2007] in order to deal explicitly with the emergence of the site. In order to do this we added variables into his list that corresponded to the location of each terraces relative to the main plaza and to the road network. A hierarchy of questions relative to location of residential and non-residential terraces was proposed. We generated answers to these questions using J48 decision trees after we compared various available algorithms and found that the J48 approach worked the best at classifying hypotheses at all levels of our hypothesis tree.;As a result, it is clear that both the Main Plaza and the road network played a major role in where terraces were located in the emergent phase of Monte Alban. However, one key question is how will this "built" environment affect future colonization of the site? In order to answer this question we added another variable to our predictor set, occupied in Ia. We then used the decision tree algorithm, J48, to predict the location of terraces in the next occupational phase, Ic.;The results show that the built environment from Phase Ia was the most important factor in terrace location in Ic. That is, terraces occupied in Ia were likely to be re-occupied in Ic. The built environment was also important in the sense that new terraces were oriented towards the main Plaza and road location.;The integration of data-mining and agent-based social learning tools allows us to infer patterns of social interaction at the site. In particular, we used decision trees rules to characterize terrace location decisions made by the early inhabitants of the major archaic urban center at Monte Alban. We then injected these rules into a socially motivated learning system based on cultural algorithms. The result was the expression of these location decisions within an inferred social fabric that provides support for two urban models.;In the process of extending previous work we have established a framework with which to mine successive phases of the site. The result of the framework is to produce decision rules that can used by virtual agents in a future simulation of site organization and emergence. It is through this simulation that we hope to learn more about the daily life and decision-making activities of these early urban dwellers.