Effect of Magnitude Differences in the Original Data on Price Forecasting. Issue 4 (December 2014)

• Record Type: Journal Article
• Authors: Zhu, Quan-Yin; Yin, Yong-Hua; Zhu, Hong-Jian; Zhou, Hong
• Is Part Of: Journal of algorithms & computational technology. Volume 8:Issue 4(2014)
• Page Start: 389
• Page End: 420
• Publication Date: 2014-12
• Subjects: Data pre-processing -- price forecasting -- magnitude differences data -- generality -- RBF neural network -- BP neural network; Computer algorithms -- Periodicals; Numerical calculations -- Periodicals; Computer algorithms; Numerical calculations; Periodicals; 518.1
• DOI: 10.1260/1748-3018.8.4.389
• Languages: English
• ISSNs: 1748-3018

Abstract

In order to ensure the stability and high accuracy on price forecasting, a novel idea is found that magnitude differences in the original data have an effect on price forecasting based on experiments. And a general data pre-processing method for the original data magnitude is put forward in this paper according to the novel idea. The new data pre-processing method that normalizes the magnitude of the original data is described in detail in this paper. The different data and models are used to verify the new idea and the generality of the proposed method. And the discrete data are the data of CNY exchange rate and agricultural product price. The different models are improved models of Back Propagation (BP) neural network and Radial Basis Function (RBF) neural network. Compared to other magnitudes, normalizing the magnitude of the original data to a smaller magnitude obtains the best result. And in this paper the smaller magnitude is the order 1. The forecasting average accuracy of the experiments done with the data of CNY exchange rate based on BP neural network and RBF neural network gets 99.14 percent and 99.27 percent. The forecasting average accuracy of the experiments completed with the data of agriculture product price based on BP neural network and RBF neural network gets 98.54 percent and 97.81 percent. Experiment demonstrates that the data pre-processing method has the great generality on discrete data and models. The data pre-processing method proposed in this paper reduces the differences between data which affect the accuracy of the prediction models to a degree. The generality of the proposed method is proved that the innovative idea and method are meaningful and useful for the price forecasting.