استعمال بعض الطرائق اللامعلمية في تقدير نموذج الانحدار الذاتي اللاخطي بوجود متغير خارجي مع تطبيق عملي == Using Some Nonparametric Methods For Estimation of Nonlinear Autoregressive Model With Exogenous Variables With Application
Author name:
علي سلمان حبیب
Supervisor name:
فراس احمد محمد
General topic:
Administration and Economics
Specific topic:
Statistics
Degree:
Doctorate
University:
University of Baghdad - Faculty Of Administration And Economics - Department Of Statistics
Language:
Arabic
University location:
Baghdad
First pages:
07T4202 - p.pdf
Abstract:
The analysis of the nonlinear time series, namely the Nonlinear Autoregressive with Exogenous Variable (NARX) model, is considered one of the complex problems.Correct order determination is very important to identify the model. Two different methods are proposed to determine the order for (NARX) model. The researcher also uses three different nonparametric methods to estimate nonlinear regression function of the model.The first proposed method to determine the order for (NARX) model was the Additive Splines Estimation to determine the correct order of the model. This method is based on Additive Property to treat the augmented ( Curse Dimensionality ) problem.The second proposed method was Cross - Validation approach leave one out, based on the kernel estimate of regression function which is directly based on data. Three different methods are used for model estimation. The first method is Smoothing Splines. The second is Artificial Neural Networks (ANN) ,and the third is the BRUTO algorithm which is an adaptive backfitting and uses Generalized Cross - Validation.For comparison purposes between the various estimation nonparametric methods and to find out the best fitting for the data, two Criterion are used : Mean Square Error (MSE) and Mean Absolute Proportional Error (MAPE), and select the best method which gives the best fitting of the data ; then applying the best method on the Electrical Loads and Temperature in Basra Governorate for the months (from May to October) in 2015. The researcher concluded that Additive Splines method play an effective role in order determination for the used model and the results show that the identified order is close to the correct order; and that the ANN is the best estimation for (NARX) model
