Study on Artificial Neural Network: A Case Study on Share Market Prediction

Abstract

Forecasting exchange rates is an important financial problem that is receiving increasing attention especially because of its difficulty and practical application. A number of techniques have emerged in the stock prediction tasks. The past researches about share forecasting usually relied on various time series models or econometric models to improve the accuracy. But, these models are limited to its linearity. Therefore, recent studies have attempted to forecast share prices using neural network models. The work combines Genetic Programming with Neural Network. It trains Neural Network with Genetic Programming and then predicts the stock price with the trained network. By learning the stock knowledge, it could find out the modes and relationship hidden in the abstract data. Exchange rates are affected by many highly correlated economic, political and even psychological factors. We propose a flexible neural tree with necessary number of hidden units and is generated initially as a flexible multi-layer feed-forward neural network evolved and also considers the approximation of sufficiently smooth multivariable function with a multilayer perceptrons. The
performance of the technique is evaluated using the stock prices.Empirical results indicate that the proposed method the flexible neural tree method is better than the forecasting methods.