الفهرس 
Title : Time Series Forecasting Models Using Machine Learning Algorithms
AbstractOnly 14 pages are availabe for public view
 |  | from | 120 |  |  |
| | | from | 120 | | |
Abstract
Time series forecasting is an active research area that plays important role in planning and decision making in several practical applications. The main task of this research area is to improve the prediction accuracy.Time series can contain both linear and nonlinear patterns. Thus, using a hybrid model can give better results in forecasting. Both linear and nonlinear parts of the time series can be modeled by this approach. While ARIMA model is used to capture the linear behavior of the time series data. Neural Network is used to model the nonlinearity in the series.This thesis proposes a novel hybrid forecasting model that models the linear and nonlinear components, which are properly decomposed from the univariate time series, by ARIMA and ANN methods, respectively, and combines the model results effectively.The proposed method, called STL-filter based hybrid ARIMA-ANN model, eliminates the important assumptions that reduce the performance results of the traditional hybrid methods. Thus, our hybrid method outperforms both individual methods and the well-accepted hybrid methods in forecasting for the time series data. Also, this study applies the methods of machine learning as modern methods of forecasting techniques and see how it could be used as an alternative method to traditional methods. The results of applying the ARIMA, and machine learning algorithms were compared through the (MSE, MAE and MAPE) results. The most important finding was that the minimum (MSE, MAE and MAPE) of the time series data used in this study using the ANN and SVM models.The results of applying the proposed hybrid method and other methods were compared through the (MSE, MAE and MAPE) results.The most important finding was that the minimum (MSE, MAE and MAPE) of the time series data used in this study using the proposed hybrid method