ANN based modeling for stock market prediction

Abstract

Analyzing stock market data and using recent developed algorithms for predicting the changes and forecasting the results is a difficult task nowaday. An efficient and faster method for selecting appropriate stock price improves market value as well as helps investors for benefits. In this paper new method implementing state of the art Artificial Neural Networks (ANNs) known as Legendre Neural Network (LENN) and Chebyshev Functional Link based Artificial Neural Network (CHFLANN) has been described and used for the analysis of stock market data. The process has been efficiently implemented in MATLAB language for representing results by evaluating accuracy and F-measure for the processed data. Prior to it the database collected is passed through a series of processes starting from normalization, expansion and then dividing it into train and test datasets to get the processed data used for implementing the learning algorithm for the stock market price evaluation and prediction purpose. In the results second order expansion systems have been used showing accuracy results as well as Mean Square Error (MSE) rates to find best method among proposed schemes. Also, our proposed models have been compared with the statistical model AutoRegressive Integrated Moving Average (ARIMA) and the basic Functional Link based Artificial Neural Network (FLANN) to analyze their performance based on the metrics like Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and R-Squared (R2).