The Classification of Complex Geographic Datasets: An Operational Comparison of Artificial Neural Network and Decision Tree Classifiers

This paper presents an overview of the functioning, control, reporting and analysis capabilities of two of the popular Artificial Intelligence techniques that are used in geocomputation as tools for classification; namely decision trees and artificial neural networks. The strengths and weaknesses of both techniques are described with respect to their operational foundation primarily, as opposed to simply contrasting results obtained. The mathematical foundation behind the two methods is not described in detail as others have fully explored these aspects.

These techniques each use a different approach to locate some minimum error or entropy solution to the standard classification problem, via strictly defined mechanisms. However, their application to geocomputational problems is often conducted in a rather haphazard fashion, which in turn can lead to unreliable results. This paper attempts to explain the classification problems encountered in geocomputation as they relate to the above techniques. Many of the datasets used are characterised by a high degree of inherent complexity, often including several layers of data, gathered according to a variety of statistical scales. Such datasets offer significant classification challenges that defy a casual approach using existing off-the-shelf tools or packages, but instead require careful consideration based on sound operational (and sometimes theoretical) knowledge of the tools to be used.