Stock Price Prediction Based On Neural Network

Stock market is closely related to people’s life and economic growth, and how to effectively and accurately predict the stock price has been one of the hot issues of people’s concern. As stock markets are influenced by political, economic and other factors, it is a very complex nonlinear dynamic system, and it becomes difficult to predict stock prices. Neural networks can theoretically fitting nonlinear transformations of arbitrary complexity, so this paper uses BP and RBF Neural network to forecast stock price.After the summary of stock price’s research background, significance, method, neural network, cuckoo theory and it’s algorithm, the problems of neural networks and its improved algorithm predicting stock prices are focused on. First of all, the traditional BP neural network’s initial connection weights and thresholds are selected randomly, which can make network fall into local optimum value easily, and aimed at the disadvantage of it, an improved algorithm is proposed to optimize network’s initial connection weights and thresholds based on Gaussian disturbance Cuckoo Search algorithm(GCS). New built PCA-GCS-BP model is used to predict Citroen logistics stock’s price. And the simulation results show that the PCA-GCS-BP model’s prediction accuracy is higher than the PCA-BP model’s. Secondly, aimed at the disadvantage of selecting center vectors parameters of RBF neural network, an improved algorithm is proposed to optimize RBF neural network’s center vectors, based on Gaussian disturbance Cuckoo Search algorithm with strong jumping out of local optimum, and it is applied to predict the stock price and stock’s rising and felling, and test samples are used to compare the forecast accuracy between RBF neural network and the tradition RBF neural network. Simulation result shows the new algorithm’s prediction accuracy is higher than the tradition RBF’s. Finally, the content of this article is summarized, and a further prospect in the future is made.