Application Of Artificial Neural Network Model In Stock Market Forecast

Since it came into being, the stock market has generated an interest amongst scientists. With investment gradually coming into public life, the stock prediction has become the focus of investors. The stock market forecast method has been developed from judgment only by analyzing prices to the modern artificial intelligent algorithms. It indicates that stock prediction method is moving in a new smarter direction. In the complex and ever-changing global economic environment, it has a wide range of practical application value that improving the accuracy of stock prediction method. In artificial intelligence technology, artificial neural network(ANN) has stronger fault-tolerant, compatibility and more processing power than other intelligent methods and has a powerful effect on predicting stock market. It is significant that applying this idea to the stock market prediction and then proposing a new algorithm depending on specific requirements.Through discussing domestic and foreign prediction methods, listing relevant evaluation indexes and analyzing the main characteristics of neural network, this paper use BP neural network and Support Vector Machine(SVM) to predict the stock market. Finally we simulate the performance of this method using MATLAB. Therefore, the following is what we study:(1) The first part puts forward the problems in stock prediction and introduces the fundamentals, structures, learning rules and drawbacks of BP neural network. And then, we improve the standard neural network by the law of additional momentum to overcome these drawbacks so that we can apply it to simulation experiments of stock market forecast system. Compared to standard neural network, the results of the experiment show that mean square error(MSE) of the improved neural network has decreased by 33% and the operation time has reduced by 50.75%. Meanwhile, the correlation coefficient has increased by 2.15%. It demonstrates that the improved neural network has better application potential than standard neural network.(2) Genetic Algorithm(GA) is a global optional algorithm, which is based on the random seeking of the theory of evolution, natural selection and the theory of inheritance. It can overcome the limitation of ANN and increase networks training and predicting speed. Therefore, this paper proposes a BP neural network model based on genetic algorithm. We apply principal component analysis(PCA) to features extraction In order to solve correlation among factors that affect the stock prediction. We use GA-BP method to forecasting after decrease the dimensions of input data. Compared to GA-BP network, the cross-validation experiments indicate that mean square error(MSE) of PCA-GA-BP network has decreased by 51% and the operation time has reduced by 21.8%. Meanwhile, the correlation coefficient has increased by 1.89%, while the number of variables has decreased by 55%. The accumulative contribution rate has increased to 99.87%. It indicates that PCA-GA-BP network has high prediction accuracy and fast calculation speed.(3) In the last part, we combine PCA and SVM to regress and predict the stock market. We optimize the parameters of SVM by using the cross-validation. It solve high dimensional input problem and has higher forecasting accuracy. Simulation experiment results show that, compared to SVM method, the mean square error(MSE) of PCA-SVM method has decreased by 70% and the operation time has reduced by 23.8%. Meanwhile, the correlation coefficient has increased by 5.8%, while the number of variables has decreased by 52.5%. The accumulative contribution rate has increased to 99.44%. It proves that the PCA-SVM forecasting model is of good feasibility and accuracy.