Forecasting in a complex environment using feature manipulating technique added in traditional forecasting system

Most forecasting systems are composed of two modules: a preprocessing module; and a learning module. In the preprocessing module, basic operations such as the removal of noise or outliners are performed. In the learning module, the knowledge contained in training data is obtained. Many forecasting systems are applicable in a simple or simplified environment and work well, yet have weak points when applied in a complex environment. That results from the characteristics of the features of training data are changed in response to training data; i.e. corresponding to the patterns of data the degrees of the influences of the features, which are subset of attributes or weighted sum of attributes, are changed. Here, the authors present a more advanced forecasting system for application in a complex environment.