Stock price pattern matching system-dynamic programming neural networks approach

The dynamic programming neural network (DNN). DNN is based on the integration of the neural and dynamic programming matching method (DP-matching). In order to find patterns similar to a specified pattern in the database, two problems must be solved. One is a nonlinear time elasticity for the patterns. This nonlinearity is normalized by DP-matching. The second is a bias which was generated by differences in names and time span. This bias is eliminated by a stock price normalization method and a neural network. A stock price pattern matching system using the DNN approach was developed on an NEC EWS4800 workstation. This system has a graphical user interface on an X-Window system. A subjective evaluation was conducted. The stock price patterns which were classified by DNN were evaluated by three chartists (human experts). High correlation was found between the similarity by DNN and the evaluation by chartists. The proposed DNN system is able to match patterns, which chartists judge as similar patterns