Based On PCA Of GA-BP Neural Network Prediction Of Stock

As people become more and more concerned about the investment ideas, there is an increasing focus on stock market in the daily life. However, the stock investment belongs to a kind of high risk and high yield investment field. Therefore stock prices prediction has greatly been under the attention of investors. Since stock market’s establishment, a lot of researchers at home and abroad began to study it and put forward many stock prices prediction methods. Based on various analyses, this paper explores using the BP neural network for building a stock prediction model. However the traditional BP algorithm has many problems, such as sensitive to the initial weights, easy to be plunged to local minimum value and a slow learning speed, etc. Therefore the prediction effect is not good. Based on these defects, this paper proposes to use Principal Component Analysis to preprocess the input variables that can reduce the input data dimension and bring down the stock price data’s noise and then use the Genetic Algorithm to optimize the network’s parameters. In the network training process, we select LM algorithm to avoid the defect that easy to fall into the local minimum problem, etc. Finally, this paper discusses the topology of the network, the principle of confirming the parameters of network. And the optimization algorithm’s feasibility used in this paper are proved by the prediction results.