An adaptive forecasting intelligent model for nonstationary time series
Iulian Năstac · 2010
The paper presents a general adaptive model for those dynamic systems that work on continuously changing environments. This interdisciplinary model is tested in the financial, genetic and technical fields. The algorithm of the model establishes how a viable structure of an artificial neural network at a previous moment of time could be retrained in an efficient manner, in order to support modifications in a complex input-output function of a real forecasting system. A remembering process from the previous learning phase is used to enhance the accuracy of the predictions. The advantage of the retraining procedure is that some relevant aspects are preserved not only from the immediate previous training phase, but also from the previous but one phase, and so on. A kind of slow forgetting process also occurs; thus it is much easier for the model to remember specific aspects of the previous training instead of an oldest one.