Access Type

Open Access Dissertation

Date of Award

January 2012

Degree Type


Degree Name



Computer Science

First Advisor

Robert G. Reynolds


In this dissertation, agent-based models of emergent ancient urban centers were constructed through the use of techniques from computational intelligence, agent-based modeling, complex systems, and data-mining of existing archaeological data from the prehistoric urban center, Monte Albán. This real world application was selected because of its importance in understanding the emergence of modern economic and political systems. Specifically, Cultural Algorithms was used to evolve models of early Monte Alban, models that can then be compared with existing models of ancient and modern urban centers.

Features of a complex system were used to help interpret the archaeological data. The analysis went from macro to meso-, to micro- and then back again. The end result is an urban morphology that is in some ways much more primitive than any of the other models proposed. However, the simplicity of the framework will allow adaptation and adjustment in the phases to follow.

First, at the macro or site level a set of decision rules for site occupation in Monte Albán Ia were generated. Monte Alban Ia is the first level of occupation associated with the site. These rules were used to suggest how the emergence of this early city fits various alternative models of urban growth. Along the way, several hypotheses were tested and conclusions were drawn as the analysis progressed. The hypotheses and the preliminary conclusions are given below.

1. The first hypothesis concerned with the nature of the attractors for the growth of the new city. The question then posed was, "Is there sufficient evidence at the macro-level to suggest that these features were important determiners of early site location here?" The resultant set of rules produced by the Decision Tree Learning algorithms generated a prediction accuracy of around 80%. Both natural and man-made features were important in forming the rules. The man-made features such as the main plaza, and road network were often the most important variable in rule formation, especially for crafts. As a result it seems clear that early settlement at the site can be described in terms of the features used in the analyses. These are features derived from the hypothesis of Marcus [Marcus, 1983].

2. The second major hypothesis related to the meso-level in terms of the neighborhood structure observed by the archaeologists during the survey. The question asked here is, "Whether there is evidence for these neighborhoods even in the early phases of settlement?" In order to test this hypothesis, the similarities and differences of the terraces in each barrio were expressed in terms of a multi-dimensional scaling approach. The results of the scaling were interpreted in terms of the micro-level classification rules produced by the Decision Trees operating at the macro or site level. It was clear from the initial scaling that there were groups or clusters of similarly classified terraces within each barrio and that the nature of these clusters varied from Barrio to Barrio. These clusters not only formed within the abstract dimensions of the analysis but were observable spatially when plotted on the ground. The level of prediction in terms of the Mardia measures was again around 80%. While this appeared to provide solid evidence for intra-barrio similarity, the clusters were generated manually.

3. The next step was to move down to the micro-level and see if there were general configurations of terraces that were held in common between barrios. These clusters of clusters would form the basis for "regions" of activity within the site, and the building blocks for the generation of a general city model. The hypothesis under consideration was, "Is there evidence for common groups or regions of activity at the micro level that are found in more than one barrio?"

The next step was to translate this information up to the regional or macro scale by looking beyond the barrios to regional clusters of clusters. The K-Means clustering had produced 29 craft cluster, 15 elite clusters, and 37 non-elite residential clusters for the 9 Barrios. They were input to the multi-dimensional scaling program and there 4 craft regions, 3 elite residential regions, and 5 non-elite residential regions. The R-Square goodness of fit for the cluster of clusters explained close to 97%, 96%, and 94% respectively for the three activities. These new clusters were produced via a hierarchical multi-dimensional scaling approach, with the result clusters interpreted manually.

The increased complexity of the scaling process suggested that rather than using K-means by itself, the approach would be guided by the Cultural Algorithm utilizing the constraints about terrace and region composition learned from the three levels analysis. That information was placed in the belief space and updated during the run of the Cultural Algorithm. The Cultural Algorithm K-Means produced four craft regions, two elite regions, and three non-elite residential regions. The results again provide support for the hypothesis.

4. Finally, at the macro-level the question posed was, "To what extent does the regional model produced by the process for the site relate to current models of urban structure and function?" It was suggested that certain craft regions were almost always associated with certain residence regions. For example, R2 and C3 were commonly found together along with R1-C4-S2. S1 remained as a single cluster, as did C1. These regions were combined and the resultant pattern of regions reflected an orientation of regional activities to those features that reflect the movement of ideas (main plaza and ceremonial centers), material goods (the road system), and the transformation of material goods (gates and reservoirs structures). The pattern, although more patchy than continuous at this point reflecting the sparseness of the early settlement, suggests an emergent sector structure. Unlike the modern version where there are no gaps between the sectors, this is more like spokes on a wheel where it is expected that in the future the spokes will expend and produce a more continuous sector structure.