ACRRFLN: Artificial Chemical Reaction of Recurrent Functional Link Networks for Improved Stock Market Prediction

Abstract

The intrinsic capability of functional link artificial neural network (FLANN) to recognize the complex nonlinear relationship present in the historical stock data made it popular and got wide applications for stock market prediction. Contrasting to multilayer neural networks, FLANN uses functional expansion units to expand the input space into higher dimensions, thus generating hyperplanes which offer better discrimination capability in the input space. The feedback properties of recurrent neural networks make them more proficient and dynamic to model nonlinear systems accurately. Artificial chemical reaction optimization (ACRO) requires less number of tuning parameters with faster convergence speed. This article develops an ACRO-based recurrent functional link neural network (RFLN) termed as ACRRFLN, in which optimal structure and parameters of a RFLN are efficiently searched by ACRO. Also two evolutionary optimization techniques, i.e., particle swarm optimization (PSO) and genetic algorithm (GA) are employed to train RFLN separately. All the models are experimented and validated on forecasting stock closing prices of five stock markets. Results from extensive simulation studies clearly reveal the outperformance of ACRRFLN over other models similarly trained. Further, the Deibold-Mariano test justifies the statistical significance of the proposed model.