Design and Implementation of Novel Artificial Neural Network Based Stock Market Forecasting System on Field-Programmable Gate Arrays

Abstract

Problem statement: Multiagent system is very proficient and has rules well-suited for financial forecast with its neural network. In financial forecasting, the approach for rules extractions is less pertinent and involves algorithms which are complex. The unsupervised network method lacks in comprehensibility and leads to ambiguity. Approach: The application of neural network technology to real-time processing of financial market analysis demands the development of a new processing structure which allows efficient hardware realization of the neural network mechanism. This study describes the realization of neural network on FPGA device for stock market forecasting system. The stock market forecasting neural network architecture consists of three layers. These are input layer with three neurons, hidden layer with two neurons and output layer with one neuron. For both output layer and hidden layer neurons, Sigmoid transfer function is used. Neuron of each layer is modelled individually using behavioural VHDL. The layers are then connected using structural VHDL. This is followed by timing analysis and circuit synthesis for the validation, functionality and performance of the designated circuit. The designated portfolio is then programmed through download cable into the FPGA chip. Results: Kuala Lumpur Stock Exchange (KLSE) index has been utilized for validating the usefulness of the completed prototype. Test on the sample of 100 data demonstrated an accuracy of 99.16% in predicting closing price of the KLSE index 10 days in advance. Conclusion: The test results are anticipated to be a higher rate of prediction for stock market analysis, thereby maintaining the high quality of supplying information in stock market business.