Learning and Decision Model Selection for a Class of Complex Adaptive Systems
Computer modeling is gaining popularity in the study of systems whose underlying processes are difficult to observe and measure directly, or their controlled experimentation is not an option. Since real-world phenomena, for instance psychological or ecological, are often hugely complicated, and the models trying to capture their essence relatively complex, validation of the models and selection among the candidates is a challenge. Furthermore, not all computer models are used merely for explanatory purposes or to test theories, but some are used to support decision making. Therefore, it is critical which model the decision makers put their confidence on. In this article I discuss a pragmatic method for selecting between classes of models that are designed to increase understanding in the most significant single factor behind the global climate change, namely human land-use. My focus is on agent-based land-use and land-cover change models, and particularly models of learning and decision making. The proposed method fills the void left by traditional statistical model selection methods that are not applicable due to the nature of the model class of interest.