A novel hybrid methodology combining back propagation neural network with rough set and its application.

In the increasingly complex world, decision makers need assistance in making informed transportation planning decisions. This article describes the application of rough set (RS) and back propagation neural network (BPNN) in the development of a novel hybrid intelligent model for forecasting the railway passenger traffic demand. The model intends to exploit the respective advantages of both RS and BPNN methods, but to avoid their respective weak points. The basic idea is to think of RS as the pre-treatment unit for BPNN that can mine knowledge from historical data. To do so, RS was used in data pre-treatment and constructing the decision table, and obtaining the minimum rule set from the decision table reduced by discretizing continuous attributes through a hybrid hierarchical k-means clustering. Input such rule set to the BPNN, trained on a sample dataset to determine parameters by using the Levenberg—Marquardt algorithm and the golden section algorithm. A hybrid intelligent forecasting model based on RS and BPNN is thus constructed and applied to forecast railway passenger traffic demand with pre-treated forecasting data. In the experiment, the railway passenger traffic data in China from 1991 to 2008 are selected as learning samples and testing samples, and the effectiveness of this method is verified in comparison to the linear recursive method. [ABSTRACT FROM AUTHOR]