Model Selection

Before using a parametric model one has to be sure that it offers a reasonable description of the system to be modeled. If a bad model structure is employed, the obtained model will also be bad, no matter how good is the parameter estimation method. There exist many possible ways of validating candidate models. This thesis focuses on one of the most common ways, i.e., the use of information criteria. First, some common information criteria are presented, and in the later chapters, various extentions and implementations are shown. An important extention, which is advocated in the thesis, is the multi-model (or model averaging) approach to model selection. This multi-model approach consists of forming a weighted sum of several candidate models, which then can be used for inference