Hybridization of the Higher Order Neural Networks with the Evolutionary Optimization Algorithms—An Application to Financial Time Series Forecasting

Abstract

AbstractMultilayer artificial neural networks have characteristics of structural complexity, computational overhead and black box visualization. In contrast, higher order neural networks (HONNs) have single layer trainable weights, fast learning properties, stronger approximation, and higher fault tolerance capability. This chapter discusses about popular HONNs such as Pi-Sigma neural network (PSNN), Sigma-Pi neural network (SPNN), and functional link artificial neural network (FLANN), their architecture, learning process, and applications to financial time series forecasting. To eradicate the limitations of gradient descent-based training, we discuss few evolutionary optimization algorithms for HONN training. The hybrid models obtained are applied to forecast financial time series. Performance analysis is carried out to establish the suitability of evolutionary optimization-based HONNs.KeywordsHigher order neural networkFinancial time series forecastingFunctional link artificial neural networkPi-sigma neural networkGenetic algorithmChemical reaction optimizationParticle swarm optimization