A Partially Connected Neural Evolutionary Network for Stock Price Index Forecasting
This paper proposes a novel partially connected neural evolutionary model (Parcone) architecture to simulate the relationship of stock and technical indicators to predict the stock price index. Different from artificial neural networks, the architecture has corrected three drawbacks: (1) connection between neurons of is random; (2) there can be more than one hidden layer; (3) evolutionary algorithm is employed to improve the learning algorithm and train weights. The more hidden knowledge stored within the historic time series data are needed in order to improve expressive ability of network. The genetically evolved weights mitigate the well-known limitations of gradient descent algorithm. In addition, the activation function is not defined by sigmoid function but sin(x). The experimental results show that Parcone can make the progress concerning the stock price index and it’s very promising to calculate the predictive percentage by simulation results of proposed evolutionary system.
KeywordsHide Layer Probabilistic Neural Network Generalize Regression Neural Network Gradient Descent Algorithm Technical Index
Unable to display preview. Download preview PDF.
- 4.Li, F., Liu, C.: Application Study of BP Neural Network on Stock Market Prediction. In: Ninth International Conference on Hybrid Intelligent Systems, Shgenyang, China, pp. 174–178 (2009)Google Scholar
- 6.Mandziuk, J.: Jaruszewicz, m.: Neuro-evolutionary approach to stock market prediction. In: Proceedings of International Joint Conference on Neural Networks, Orlando, Florida, USA, pp. 12–17 (August 2007)Google Scholar
- 10.Montana, D.: A Weighted Probabilistic Neural Network. In: Advances in Neural Information Processing Systems, pp. 1110–1117 (1992)Google Scholar