An Empirical Study of Integrating Independent Component Analysis and Combined Forecasting for Time Series Data

• Main Authors: Guan-Yu Zhou, 周冠宇
• Other Authors: Feng-Jenq Lin
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/33dxaf

碩士 === 國立宜蘭大學 === 應用經濟與管理學系應用經濟學碩士班 === 102 === In this study, we use several forecasting methods via Independent Component Analysis Whitening process to get eigenvalues of components, and then we integrate concept of eigenvalue weight with Combined Forecasting. Finally, we create a new Combined Forecasting method. In research we consider all the possible combinations of weighted average between weight values and forecast values. In order to make the operation more quickly, so we according to the actual values and predicted values of the datas, and then we use their historical data to figure out the Mean Absolute Error（MAE）or the Root Mean Square Error（RMSE）as the basis of decision weights that are used to allocate weights and create the direct allocating method whereby achieve the purpose of simplification the all possible method. In empirical analysis, we use the four test cases and four actual cases about the traffic statistics to predict performance evaluation and make comparison. Experimental results show direct allocating method and all possible method can outperform single forecasting method. They can outperform arithmetic average method and inverse variance method for statistics of traffic prediction. This finding can be used as the statistics of traffic prediction method for reference. Finally, we collate case study results, so we suggest that research can also apply to other data in the areas of forecasting. For example, price statistics, labor statistics, finance statistics, population statistics, environmental protection statistics, etc.