Search

Aladag, Cagdas Hakan/0000-0002-3953-7601; Egrioglu, Erol/0000-0003-4301-4149 WOS: 000263150600012 The only suggestions given in the literature for determining the architecture of neural networks are based on observations, and a simulation study to determine the architecture has not yet been reported. Based on the results of the simulation study described in this paper, a new architecture selection strategy is proposed and shown to work well. It is noted that although in some studies the period of a seasonal time series has been taken as the number of inputs of the neural network model, it is found in this study that the period of a seasonal time series is not a parameter in determining the number of inputs.

KADILAR, CEM/0000-0003-4950-9660; Aladag, Cagdas Hakan/0000-0002-3953-7601; Egrioglu, Erol/0000-0003-4301-4149 WOS: 000277172100010 Artificial neural networks can successfully model time series in real life. Because of their success, they have been widely used in various fields of application. In this paper, artificial neural networks are used to model brain wave data which has been recorded during the Wisconsin Card Sorting Test. The forecasting performances of different artificial neural network models, such as feed forward and recurrent neural networks, using both linear and nonlinear activation functions in the output neuron, are examined. As a result of the analysis, it is found that artificial neural networks model the data successfully and all the models employed produce very accurate forecasts.

Aladag, Cagdas Hakan/0000-0002-3953-7601; Egrioglu, Erol/0000-0003-4301-4149 WOS: 000312412800005 Fuzzy time series forecasting methods, which have been widely studied in recent years, do not require constraints as found in conventional approaches. On the other hand, most of the time series encountered in real life should be considered as fuzzy time series due to the vagueness that they contain. Although numerous methods have been proposed for the analysis of time series in the literature, these methods fail to forecast seasonal fuzzy time series. The limited number of seasonal fuzzy time series methods consider only the fuzzy set having the highest membership value, rather than the membership value of observations belonging to each fuzzy set. This is contrary to fuzzy set theory and causes information loss, thus affecting forecasting performance negatively. In this study, a new seasonal fuzzy time series method which considers the membership value of the observations belonging to each set in both forecasting fuzzy relations and in the defuzzification step is proposed. The proposed method is applied to a real seasonal time series.

Aladag, Cagdas Hakan/0000-0002-3953-7601; Egrioglu, Erol/0000-0003-4301-4149 WOS: 000312412800001 When fuzzy time series include a seasonal component, conventional fuzzy time series models are not sufficient. For such fuzzy time series, lagged variables which are around the period of the time series should also be included in the model. Determining the lagged variables which will be in the forecasting model is a vital issue. Also, defining fuzzy relations is another important issue in the fuzzy time series approach. When the number of fuzzy lagged variables is large, using artificial neural networks to define fuzzy relations makes the operations easier and increases the forecasting accuracy. In this study, in order to deal with the problem of determining the lagged variables, and defining the fuzzy relations; a novel seasonal fuzzy time series approach based on SARIMA and the multiplicative neuron model is proposed. In the proposed method, the SARIMA method is exploited to choose the fuzzy lagged variables and multiplicative neuron model is employed to establish the fuzzy relations. To show the applicability of the proposed method, it is applied to the invoice sum accrued to health service providers. For comparison, the data is also analyzed with other fuzzy time series approaches in the literature. It is observed that the proposed method has the best forecasting accuracy with respect to other methods.