Dendritic Neuron Model Neural Network Trained by Modified Particle Swarm Optimization for Time-Series Forecasting

Abstract

Different types of artificial neural networks (NNs), such as nonprobabilistic and computation-based time-series forecasting tools, are widely and successfully used in the time-series literature. Whereas some of them use an additive aggregation function, others use a multiplicative aggregation function in the structure of their neuron models. In particular, recently proposed sigma-pi NNs and dendritic NNs have additional and multiplicative neuron models. This study aims to take advantage of the dendritic neuron model neural network (DNM-NN) in forecasting and hence uses the DNM-NN trained by a modified particle swarm optimization as the main contribution of the study optimization in time-series forecasting to improve the forecasting accuracy. To evaluate the forecasting performance of the DNM-NN, the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) was analyzed, and the obtained results were discussed together with the results produced by other time-series forecasting models, including traditional, fuzzy-based, and computational-based models.