A resolution to stock price prediction by developing ANN-based models using PCA

Abstract

The application of artificial neural network (ANN) has become quite ubiquitous in numerous disciplines with different motivations and approaches. One of the most contemporary implementations accounts it for stock price behavior analysis and forecasting. The stochastic behavior of stock market follows numerous factors to determine the price vicissitudes such as GDP, supply and demand, political influences, finance, and many more. In this paper, we have considered two ANN techniques, viz., backpropagation-based neural network (BPNN) and radial basis function network (RBFN), first, without principal component analysis (PCA), and further modified the model with PCA, to execute financial time series forecasting for the next 5 days (which can also be extended for some other number of days) by accepting the input as historical data on the sliding window basis. Moreover, the empirical research is conducted to verify the forecasting impact on the stock prices for oil and natural gas sector in India with the developed model, and subsequently a comparison study has also been performed for the effectiveness of the two models without and with PCA, on the basis of mean square percentage error.