Fuzzy Time Series Forecasting Model Using Particle Swarm Optimization and Neural Network

In recent years, there are several ongoing efforts to develop models for forecasting fuzzy time series using classical or artificial intelligence (AI) techniques in different application areas. A major challenge lying with the fuzzy time series forecasting model is efficient partitioning of data. It has significant effect on forecasting accuracy. Proposed work overcomes the difficulty of searching appropriate interval length for partitioning the data. In this study, a hybrid model using particle swarm optimization and backpropagation neural network (BPNN) is applied for forecasting fuzzy time series. Particle swarm optimization (PSO) searches for optimal partitioning of data, and weights of neural network are adjusted using gradient descent technique. The neural network takes fuzzy membership values as input, and every particle represents a set of boundaries between two adjacent intervals. This hybrid procedure is iterated until stopping condition is reached. The experiment is carried out on standard datasets, and results are compared with related models including neuro-fuzzy models applied on the same dataset. Proposed idea shows best performance in terms of accuracy in prediction.