The application of Genetic Algorithm-Radial Basis Function (GA-RBF) Neural Network in stock forecasting

Abstract

According to the shortage that only historical data are made use of in the previous researches on stock forecast, a new idea of multi-input stock forecasting integrating various outer impact factors such as Dow Jones Index, Nikkei Index and Hang Seng Index etc. was presented. To avoid the local convergence of BP Neural Network, Radial Basis Function Neural Network (RBF) was selected and Genetic Algorithm (GA) was adopted for parameter optimization of RBF, and then forecasting was carried out by making use of the GA-RBF network obtained after optimization. This approach has good generalization capability and learning speed, which overcomes the shortages in BP network and solves the problem that a unified standard is lacked for RBF network parameter selection. The experiment results indicate that the approach of this paper can reflect the impact factors more complete and thus works better.