2A-2 An Application of Web-based Hybrid Approach of Neural Networks and the Linear Regression Model to the Water Supply forecast in Tokyo District 23

概要

論文の詳細を見る
Linear regression analysis and artificial neural networks are two useful data analysis techniques and both possessing distinctive advantages namely "simple and clear interpretation" for the former and "sufficient forecast accuracy" for the latter. Our proposed hybrid approach is an attempt to combine the strength of these two well-received methods with a view to producing accurate forecast with high visibility of the statistical correlation between the variables under consideration. This does not need prior knowledge of the characteristics or number of structural changes but only a certain set of input variables, which can describe the property of the incompleteness of the data. The key idea of the hybrid approach to effective data editing is to take the output of appropriate hidden units in the neural networks as additional input variables of the linear regression analysis. This satisfies both the flexibility of the neural networks for approximating input variation of the regression function and the clarity of the interpretation through the coefficients of linear regression representing the ordinary variation of the regression function. The hybrid approach also provides a structured way to select the input variables from the potential candidates. Previously, we have applied this specific hybrid approach for editing of impure financial data as well as modelling time-series construction price index and cross-section construction price. We also developed the web-based system to automate the calculation.
日本計算機統計学会の論文
2008-05-22

2A-2 An Application of Web-based Hybrid Approach of Neural Networks and the Linear Regression Model to the Water Supply forecast in Tokyo District 23

スポンサーリンク

概要

著者

関連論文

スポンサーリンク