Forecasting nonlinear time series with a hybrid methodology

Applied Mathematics Letters (Impact Factor: 1.48). 09/2009; 22(9):1467-1470. DOI: 10.1016/j.aml.2009.02.006
Source: DBLP

ABSTRACT a b s t r a c t In recent years, artificial neural networks (ANNs) have been used for forecasting in time series in the literature. Although it is possible to model both linear and nonlinear structures in time series by using ANNs, they are not able to handle both structures equally well. Therefore, the hybrid methodology combining ARIMA and ANN models have been used in the literature. In this study, a new hybrid approach combining Elman's Recurrent Neural Networks (ERNN) and ARIMA models is proposed. The proposed hybrid approach is applied to Canadian Lynx data and it is found that the proposed approach has the best forecasting accuracy.

Download full-text


Available from: Cem Kadilar, Jul 02, 2015
  • Source
    Dyna (Medellin, Colombia) 08/2013; 80(180):4-8. · 0.22 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Muchas series de tiempo con tendencia y ciclos estacionales son exitosamente modeladas y pronosticadas usando el modelo airline de Box y Jenkins; sin embargo, la presencia de no linealidades en los datos son despreciadas por este modelo. En este artículo, se propone una nueva versión no lineal del modelo airline; para esto, se reemplaza la componente lineal de promedios móviles por un perceptrón multicapa. El modelo propuesto es usado para pronosticar dos series de tiempo benchmark; se encontró que el modelo propuesto es capaz de pronosticar las series de tiempo con mayor precisión que otras aproximaciones tradicionales. Many time series with trend and seasonal pattern are successfully modeled and forecasted by the airline model of Box and Jenkins; however, this model neglects the presence of nonlinearity on data. In this paper, we propose a new nonlinear version of the airline model; for this, we replace the moving average linear component by a multilayer perceptron neural network. The proposed model is used for forecasting two benchmark time series; we found that the proposed model is able to forecast the time series with more accuracy that other traditional approaches.
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Artificial neural networks (ANNs) have been applied to time series forecasting. Genetic algorithm (GA) can be used as an optimization search scheme to determine the near optimal architecture and parameters of a neural network, as well. In this study a rich evolutionary connectionist model is proposed, in which GA is used to determine the optimum number of input and hidden nodes of a feedforward neural network, the optimum slope of nodes' activation function and the optimum values of learning rates and momentum coefficients. Empirical results on foreign exchange rate prediction indicate that the proposed hybrid model exhibits effectively improved accuracy, when is compared with some other time series forecasting models.