人工神經元網路 (ANN) 模型廣泛應用於資料科學領域，特別是它具有良好預測非線性結構資料的能力。本文感興趣的第一個研究議題為: 針對各種不同時間序列資料結構下，使用統計領域的 Seasonal ARIMA (SARIMA) 模型與資訊工程領域的 ANN 模型，在一步及多步預測的效果優劣比較。第二個研究議題為是否能透過 ANN 模型的結構，擷取或辨識非線性結構資料的特徵 ? 本文提供一個探索結構特徵的方法。
我們利用模擬數據與 實際資料進行 SARIMA 跟 ANN 的比較。在模擬試驗中，我們檢視不同程度之非線性隨機過程，在不同樣本數下的預測表現，希望探究ANN 模式在時序資料預測的應用成效。結果顯示當樣本數足夠多時，ANN對非線性 threshold AR 模式確實具有較佳的預測能力，同時也能透過觀察內部結構 (hidden layer) 學習資料特徵。本論文並綜合實例資料及模擬資料的分析結果以建議 SARIMA 及 ANN 模型適合使用時機。

Artificial Neural Network (ANN) model has been widely applied in data science analysis, since it has good performance on predicting data with nonlinearity features. The first topic this thesis interested in is : What is the 1-step-ahead and multi-step-ahead forecast performance using Seasonal ARIMA (SARIMA) in the statistics literature and ANN in the information engineering literature. The second topic is : Is it available to extract or identify the nonlinearity feature from data via ANN fitting procedure? This thesis provides some exploration in this direction.
We use simulation data and real data to compare SARIMA and ANN forecast performance. In simulation study, nonlinear random processes are generated with different degrees of nonlinearity with different sample size, and compare SARIMA and ANN forecast performance. We find that ANN model is effectiveness when it applied in time series forecasting. The result shows when sample size is large, ANN has better forecast performance. We can also learn feature of data through observing the ANN hidden layer structure. This thesis advises the appropriate scenarios of using SARIMA and ANN through the analysis results of simulation data and real data.

1.Anderson, B. and Moore, J. B. (1979). Optimal Filtering, Prentice-Hall, EnglewoodCliffs, New Jersie, 1st edition.
2.Box, G. and Jenkins, G. M. (1970). Time Series Analysis: Forecasting and Control,Holden-Day, San Francisco, 1st edition.
3.Gurney, K. (1997). An introduction to neural networks, UCL Press, London, 1st edition.
4.Hamzacebi, C. (2008). Improving artificial neural networks performance in seasonal time series forecasting, Information Sciences, 4550-4559.
5.Hochreiter, S.(1997). Long Short-Term Memory, Neural Computation, 9,
1735-1780.
6.Hyndman, R. and Khandakar, Y. (2008). Automatic time series forecasting: The forecast package for R, Journal of Statistical Software, 26, ISSUE 3, 1-22.
7.Kendall, M. and Stuart, A. (1983). The Advanced Theory of Statistics, Grin, 3,New York, 1st edition.
8.Livera, A. and Hyndman, R. (2011). Forecasting Time Series With Complex Seasonal Patterns Using Exponential Smoothing, Journal of the American Statistical Association, 106, 1513-1527.
9.Lapedes, A. and Farber, R. (1988). How Neural Networks Works, in Neural Information Processing Systems, American Institute of Physics, 442-456.
10.Schuster, M. and Paliwal, K.K. (1997). Bidirectional recurrent neural networks,IEEE Transaction on signal processing, 2673-2681.
11.Zhang, G.P. (1998). Forecasting with artificial neural networks : The state of the art, International Journal of Forecasting, 14, 35-62.
12.Zhang, G.P. (2003). Time series forecasting using a hybrid ARIMA and neural network model, Neurocomputing, 50, 159-175.