Improvement of ANN Models via Data Envelopment Analysis for Stock Prices Forecasting

Abstract

The implementation of artificial neural network techniques has become quite prevalent in the field of nonlinear data modelling and forecasting in this era. The only application of ANN models for model fitting may not be sufficient for close and satisfactory performances; hence the researchers are adopting hybrid models of ANN with different statistical and machine learning approaches such as support vector machines, particle sworm optimization, principal component analysis, etc. We have also developed a hybrid model in this paper with ANN and data envelopment analysis (DEA) techniques for stock prices forecasting in share market. The efficient decision making units have been selected with help of DEA approach and provided it as input to the Lavenberg-Marquardt technique based ANN model in sliding window manner. Further a closed performance of our hybrid model has been achieved by carrying out our experimentation with different number of nodes in the hidden layer of ANN model. Since the prices of stocks follow numerous factors such as demand and supply, political environments, economy and finance, buy and sell, etc., the historical prices for stocks may be convenient for the further forecasting.