Local weighted approach to time series forecasting
Ventsislav Georgiev Nikolov · 2011
In this paper an approach is proposed for associating priorities to the data according to their actuality and using of local neural network forecasting method. For this purpose modified learning rules are derived that lead to shifting of the prototype vectors in the self-organizing map and weighted training in the multilayer perceptron. This method is generally applicable to the forecasting of long and complex time series.