The first step in interrogating the data for a possible relationship is the selection of a limited amount of data, called the training data, from which a model will be developed. The model is an idealized relationship, involving a number of variables, that is suggested by initial examination of the training data or by practical observations. Many different kinds of models are in use, having been drawn from different disciplines. Predictive analytics is essentially a statistical process in that the results obtained are not precise but are expressed in terms of probability. Thus, levels of reliability in terms of confidence limits are a feature. The various statistical methods that we have discussed in previous chapters have their use in setting up proposed models. In addition, techniques from studies of machine learning, artificial intelligence, and neural networks are in use. The development of new and improved models is an active area of research. The following sections are intended to give an indication of the kinds of models that are used and the way in which they work.